WO2023175910A1 - Decision support system, decision support method, and recording medium - Google Patents

Decision support system, decision support method, and recording medium Download PDF

Info

Publication number
WO2023175910A1
WO2023175910A1 PCT/JP2022/012639 JP2022012639W WO2023175910A1 WO 2023175910 A1 WO2023175910 A1 WO 2023175910A1 JP 2022012639 W JP2022012639 W JP 2022012639W WO 2023175910 A1 WO2023175910 A1 WO 2023175910A1
Authority
WO
WIPO (PCT)
Prior art keywords
objective function
race
optimal solution
support system
decision
Prior art date
Application number
PCT/JP2022/012639
Other languages
French (fr)
Japanese (ja)
Inventor
文秀 瀧本
宗裕 橋本
慎之介 西本
Original Assignee
日本電気株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電気株式会社 filed Critical 日本電気株式会社
Priority to PCT/JP2022/012639 priority Critical patent/WO2023175910A1/en
Publication of WO2023175910A1 publication Critical patent/WO2023175910A1/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/34Betting or bookmaking, e.g. Internet betting

Definitions

  • the present invention relates to a decision support system and the like.
  • the learning device of Patent Document 1 generates an objective function for determining an optimal solution by inverse reinforcement learning based on decision history data.
  • the program programming program in Patent Document 2 uses a learning model to organize athletes who will participate in a race.
  • the learning device of Patent Document 1 and the program programming program of Patent Document 2 may not be able to present optimal proposals in decision-making regarding race results of publicly managed competitions.
  • the purpose is to provide a decision-making support system, etc. that can easily make decisions regarding the race results of publicly managed competitions.
  • the decision support system of the present invention includes an acquisition means for acquiring information regarding races in publicly managed competitions, and a decision support system that is generated in advance by inverse reinforcement learning based on the decision history regarding prediction of race results in publicly managed competitions.
  • the decision-making support method of the present invention acquires information regarding races in publicly managed competitions, and uses objective functions generated in advance by inverse reinforcement learning based on decision history regarding prediction of race results in publicly managed competitions, and the acquired race results in publicly managed competitions.
  • the optimal solution for predicting the race result is determined using the information regarding the race result, and information regarding the prediction of the race result is output based on the determined optimal solution.
  • the recording medium of the present invention includes a process for acquiring information regarding races in publicly managed competitions, an objective function generated in advance by inverse reinforcement learning based on a decision history regarding prediction of race results in publicly managed competitions, and an acquired race result in publicly managed competitions.
  • a non-decision support program that causes a computer to execute the process of determining the optimal solution for predicting the race result using information about the race result, and the process of outputting information regarding the prediction of the race result based on the determined optimal solution. Record temporarily.
  • FIG. 1 is a diagram showing an example of a configuration of a first embodiment of the present invention.
  • FIG. 1 is a diagram showing an example of the configuration of a decision support system according to a first embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. It is a figure showing an example of a display screen in a 1st embodiment of the present invention. FIG.
  • FIG. 3 is a diagram showing an example of the operation flow of the decision support system according to the first embodiment of the present invention.
  • FIG. 3 is a diagram showing an example of the operation flow of the decision support system according to the first embodiment of the present invention. It is a figure showing an example of composition of a 2nd embodiment of the present invention. It is a figure showing an example of composition of a decision support system of a 2nd embodiment of the present invention. It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. It is a figure showing an example of a display screen in a 2nd embodiment of the present invention.
  • FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention.
  • FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention.
  • FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention.
  • FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention.
  • FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention.
  • It is a figure showing an example of a display screen in a 2nd embodiment of the present invention.
  • It is a figure showing an example of a display screen in a 2nd embodiment of the present invention.
  • FIG. 1 is a diagram showing an example of an information processing system according to this embodiment.
  • the information processing system includes a decision support system 10, a user terminal device 20, and an information management server 30.
  • the decision support system 10 is connected to a user terminal device 20 via a network. Further, the decision support system 10 is connected to an information management server 30 via a network.
  • the decision support system 10 is a system that supports decision making regarding the race results of publicly managed competitions.
  • An example of a publicly managed game is horse racing.
  • Publicly managed competitions may be bicycle races, boat races, or auto races. Examples of publicly managed competitions are not limited to those mentioned above, and the type of competition does not matter as long as it is a competition held by a public institution for gambling purposes.
  • Decision-making regarding race results in publicly managed competitions is, for example, determining the combination of voting tickets when purchasing voting tickets for races in publicly managed competitions.
  • the decision regarding the race result of the publicly managed competition may be a decision on predicting the finish order of the race or a decision on the race for which voting tickets are to be purchased.
  • the combination of voting tickets is determined by determining, for example, when purchasing voting tickets, what kind of horse numbers are used for two consecutive wins (horse racing) and frame numbers are two consecutive winning combinations (fraction). The key is to decide whether to buy a combination.
  • the combination of voting tickets is not limited to the above example.
  • examples of decision-making regarding race results in publicly managed competitions are not limited to determining the combination of voting tickets.
  • the decision support system 10 uses, for example, a pre-generated objective function to determine the optimal solution for predicting race results.
  • the objective function is generated by inverse reinforcement learning based on the history of decision-making regarding prediction of race results in public competitions.
  • the optimal solution in predicting race results is also called an optimization result.
  • the decision support system 10 then outputs information regarding the race results to the user terminal device 20 based on the determined optimal solution.
  • the objective function is a function used to determine the optimal solution for predicting race results.
  • the feature amounts correspond to explanatory variables in the objective function.
  • the feature amount is, for example, information regarding a race that can influence decision making.
  • the feature amounts are, for example, pedigree and race distance.
  • the weighting coefficient of the feature amount corresponds to the coefficient of the explanatory variable in the objective function.
  • the weighting coefficient of the feature amount is an index indicating how much importance is given to information regarding the race in decision making.
  • the combination of the feature quantities included in the objective function generated based on the decision-making history regarding prediction of race results in public competitions and the weighting coefficients of the feature quantities is also called intention.
  • the combination of the feature amounts included in the objective function and the weighting coefficients of the feature amounts reflects the decision-making intention of the person whose decision history is used when generating the objective function.
  • the intention is explained using natural language (including words and sentences) according to decision-making tendencies, for example.
  • a feature amount having a weighting coefficient greater than or equal to a predetermined value in the objective function may be used as a word corresponding to a decision-making tendency.
  • leg quality is set as a word according to the tendency of decision making.
  • the objective function is generated, for example, reflecting the intention of an expert.
  • a plurality of objective functions may be generated, for example, reflecting the intentions of each expert.
  • the decision support system 10 may determine the optimal solution for predicting the race result using any one of the plurality of objective functions generated for each intention.
  • the decision-making history regarding the prediction of the race result of a publicly managed competition includes status data regarding the publicly managed competition and behavioral data regarding the publicly managed competition.
  • the status data regarding publicly managed competitions is, for example, information regarding races in publicly managed competitions.
  • Information regarding races in public competitions includes, for example, race conditions, odds, and attributes of athletes participating in the races.
  • the competition body is the main body that participates in the race.
  • the attributes of the competition objects are information about each competition object participating in the race.
  • the publicly managed competition is horse racing
  • the competition object is a horse.
  • the attributes of the sport also include information regarding the jockey.
  • the publicly managed competition is a bicycle race, a boat race, or an auto race
  • the competing objects are athletes.
  • the attributes of the competition object may include information about a bicycle, a boat, or a motorcycle. Information regarding races is not limited to the above example.
  • Behavioral data related to publicly managed competitions is, for example, the result of a decision made intentionally by a person regarding a race in a publicly managed competition.
  • a decision made with intention is, for example, a decision made by a person making the decision based on a policy that the person has.
  • the decision history includes information about a certain race in the past and the decision made by an expert regarding the combination of voting tickets for that race. It's a decision.
  • the decision regarding the combination of voting tickets is, for example, the determination of the purchasing pattern of voting tickets.
  • the voting ticket purchase pattern is, for example, a combination of voting tickets to be purchased.
  • the voting ticket purchase pattern may be a combination of voting tickets and a purchase amount for each voting ticket.
  • the information regarding a certain race in the past is status data.
  • the combination of voting tickets determined by the expert for the race is behavioral data.
  • the decision-making history may be data on decisions made by experts or commentators.
  • the decision-making support system 10 generates an objective function by inverse reinforcement learning, for example, based on the decision-making history regarding prediction of race results in publicly managed competitions in past races.
  • the generated objective function is used to determine the optimal solution in predicting race results.
  • the decision support system 10 generates an objective function by determining weighting coefficients of features included in the objective function, for example, by inverse reinforcement learning using race information and decision history in past races. .
  • the weighting coefficient of the feature included in the objective function is the coefficient of the explanatory variable corresponding to the feature in the objective function.
  • the decision support system 10 may use an objective function generated outside the decision support system 10 to determine the optimal solution for predicting the race result.
  • the user terminal device 20 is, for example, a terminal device owned by a person who uses the optimal solution in predicting the race result determined by the decision support system 10.
  • a person who uses the optimal solution in predicting a race result is, for example, a person who purchases a voting ticket for a race.
  • the person who utilizes the optimal solution in predicting the race result may be the person in charge of devising the organization of the race. Further, the person who uses the optimal solution in predicting the race result may be a reporter or a commentator.
  • the person who uses the optimal solution in predicting the race result is not limited to the above example.
  • the information management server 30 is, for example, a server that holds information regarding races in publicly managed competitions.
  • the information regarding the race is, for example, the conditions of the race and the attributes of the athletes participating in the race.
  • the competition body is the main body that participates in the race.
  • the attributes of the competition objects are information about each competition object participating in the race.
  • the publicly managed competition is horse racing
  • the competition object is a horse.
  • the attributes of the sport also include information regarding the jockey.
  • the publicly managed competition is a bicycle race, a boat race, or an auto race
  • the competing objects are athletes.
  • the attributes of the competition object may include information about a bicycle, a boat, or a motorcycle. Information regarding races is not limited to the above example.
  • the decision support system 10 acquires information regarding races in public competitions from the information management server 30. Then, the decision support system 10 determines the optimal solution for predicting the race result using the information regarding the race of the publicly managed competition acquired from the information management server 30 and the objective function. The decision support system 10 then outputs the optimal solution for predicting the race result to the user terminal device 20.
  • the decision support system 10 may acquire information regarding races in public competitions from a plurality of information management servers 30. Further, the decision support system 10 may acquire information regarding races in public competitions from the user terminal device 20.
  • the decision support system 10 may output the optimal solution for predicting the race result to the plurality of user terminal devices 20.
  • the decision support system 10 may output the optimal solution for predicting the race result to the user terminal devices 20 used by a plurality of users.
  • the number of user terminal devices 20 and information management servers 30 may be set as appropriate.
  • FIG. 2 is a diagram showing an example of the configuration of the decision support system 10.
  • the decision support system 10 includes an acquisition section 11 , a determination section 12 , an output section 13 , a generation section 14 , and a storage section 15 .
  • the acquisition unit 11 acquires information regarding races in publicly managed competitions.
  • Information regarding races in public competitions is, for example, information that can be related to determining the optimal solution in predicting race results.
  • the acquisition unit 11 acquires, for example, race conditions, odds, and attributes of competition objects as information regarding races in publicly managed competitions.
  • the acquisition unit 11 acquires odds for each type of voting ticket.
  • the race conditions include, for example, information regarding the stadium. Further, the race conditions may include race setting conditions and conditions that the runners must satisfy.
  • the race conditions include, for example, at least one of distance, race track type, race horse conditions, weight, race rating, race track, weather, race track condition, number of race horses, and running direction.
  • the type of horse track is, for example, grass, dirt, or obstacles.
  • the conditions for a horse to run are defined by, for example, the age and sex of the horse.
  • the condition of the horse track is, for example, information indicating the amount of water contained in the horse track.
  • the running direction is, for example, information indicating whether the race is run counterclockwise or clockwise. Race conditions when the publicly managed competition is horse racing are not limited to the above example.
  • the attributes of the competition are the attributes of the horses running.
  • the attributes of the runners are information about each runner.
  • the attributes of the running horses include, for example, lane number, slot number, odds, age, gender, weight, weight change, blood data, muscle mass, training status, health condition, rest history, race participation history, weight load, leg quality, At least one of race record, sire horse, dam horse, owner, stable, trainer, and producer.
  • the lane number may be a horse number.
  • the training situation is, for example, time and time change for each distance during training.
  • the foot quality is set, for example, by classification of a runaway horse, a leading horse, a lead horse, or a chasing horse.
  • the attributes of the running horses may include the race records of the sire horse and dam horse.
  • race results include, for example, race conditions in past races, attributes of horses running in races, prize money earned, and race developments.
  • Lace development includes, for example, positioning and difference in placement. The positioning is, for example, the ranking and time in each section when the entire section of the race is divided into predetermined distances.
  • the difference in finish is, for example, the time difference between a horse that is higher in the ranking or a horse that is lower in the ranking.
  • the attributes of the running horses are not limited to the above example.
  • the race conditions include, for example, at least one of the race track, race distance, and weather.
  • the attributes of the competition body include at least one of the athlete's height, weight, age, leg quality, odds, and match record.
  • examples of race conditions and attributes of the race bodies are not limited to the above.
  • the race conditions include, for example, at least one of the race track, race distance, and weather.
  • the attributes of the competition body include at least one of the athlete's height, weight, age, rank, odds, and match record.
  • examples of race conditions and attributes of competition objects are not limited to the above.
  • the race conditions include, for example, at least one of the race track, the presence or absence of a handicap, and the weather.
  • the attributes of the athlete are at least one of the athlete's height, weight, age, affiliation, class, odds, and match record.
  • the attributes of the competition object may include start display information. Examples of race conditions and attributes of race objects when the publicly managed competition is an auto race are not limited to the above.
  • the acquisition unit 11 may acquire from the user terminal device 20 the selection result of the objective function used to determine the optimal solution.
  • the selection result of the objective function may be acquired from the user terminal device 20, for example, by inputting the selection of the objective function to the user terminal device 20 through the operation of a person who uses the optimal solution in predicting the race result. Further, the acquisition unit 11 may acquire from the user terminal device 20 a selection of display items that is input to the user terminal device 20 by an operation of a person who uses the optimal solution in predicting the race result.
  • the acquisition unit 11 acquires the decision history regarding races in publicly managed competitions as data for generating the objective function. You may.
  • the acquisition unit 11 stores, for example, the acquired decision-making history regarding races in public competitions in the storage unit 15.
  • the determining unit 12 determines the optimal solution for predicting race results using the objective function and information regarding races in publicly managed competitions.
  • the objective function is generated in advance using inverse reinforcement learning based on the decision-making history related to predicting race results in publicly managed competitions.
  • the determining unit 12 determines an optimal solution for predicting race results by solving an optimization problem using, for example, an objective function and information regarding races in publicly managed competitions.
  • the determining unit 12 may extract the reason for determining the optimal solution in predicting the race result. For example, the determining unit 12 extracts a feature quantity that has a larger influence on determining the optimal solution than other items as the reason for determining the optimal solution. A feature that has a greater influence on determining the optimal solution than other items is, for example, a feature that has a larger weighting coefficient than the other features. Further, the determining unit 12 outputs a feature amount whose degree of influence on determining the optimal solution is equal to or higher than a preset standard as a reason for determining the optimal solution.
  • the determining unit 12 may output information regarding intentions that have a greater influence on determining the optimal solution than other intentions as the reason for determining the optimal solution.
  • the information regarding the intention is, for example, a mathematical formula corresponding to the intention or a word indicating the intention.
  • the determining unit 12 outputs information regarding intentions whose influence on determining the optimal solution is greater than a preset standard as the reason for determining the optimal solution. Good too.
  • the determining unit 12 may determine the optimal solution for predicting the race result using any one of the plurality of objective functions.
  • the determining unit 12 determines the optimization result using, for example, an objective function according to the attribute of the person who uses the optimal solution in predicting the race result.
  • the attribute of the person who uses the optimal solution in predicting the race result is, for example, the position of the person who uses the optimal solution in predicting the race result.
  • the attributes of the person who uses the optimal solution in predicting the race result are set, for example, by classification of voting ticket purchaser, reporter, commentator, and race organizer.
  • the attributes of the person who uses the optimal solution in predicting the race result may be set for beginners, intermediates, and experts.
  • the attributes of the person who uses the optimal solution are not limited to the above.
  • the determining unit 12 may determine the optimal solution for predicting the race result using an objective function generated for each person whose decision-making history is used.
  • the determining unit 12 uses, for example, either model A that uses an objective function generated based on commentator A's decision-making history, or model B that uses an objective function generated based on commentator B's decision-making history, An optimal solution for predicting race results may be determined.
  • the determining unit 12 uses, for example, an objective function to determine a combination of voting tickets that will result in a higher payout amount. For example, the determining unit 12 determines a combination of voting tickets such that the refund amount is higher than the purchase amount. For example, the determining unit 12 determines a combination of voting tickets such that the refund amount is higher than the purchase amount, using the budget for purchasing voting tickets as a constraint condition.
  • the determining unit 12 may determine a combination of voting tickets for a plurality of races as the optimal solution. In order to obtain a high return, the determining unit 12 may determine a combination of voting tickets from among voting tickets with odds equal to or higher than a reference value as an optimal solution. Furthermore, in order to prevent speculative purchases, the determining unit 12 may determine a combination of voting tickets from among voting tickets whose odds are less than a standard as an optimal solution.
  • the determining unit 12 may determine the race for which a voting ticket is to be purchased as the optimal solution.
  • the determination unit 12 may determine a race for which a voting ticket is to be purchased from among races that satisfy the constraint conditions.
  • the determining unit 12 determines, for example, a race for which a voting ticket is to be purchased from among races within a certain period of time after the person using the optimal solution earns income.
  • the determining unit 12 may determine the order of race finish as the optimal solution.
  • the prediction of the race result determined as the optimal solution is not limited to the above example.
  • the output unit 13 outputs information regarding prediction of the race result based on the optimal solution determined by the determination unit 12.
  • the information regarding the prediction of the race result is, for example, at least one of the optimal solution determined by the determining unit 12, the reason for determining the optimal solution, and information generated using the determined optimal solution.
  • Information regarding prediction of race results is not limited to the above.
  • the optimal solution is a combination of voting tickets to be purchased
  • the information generated using the determined optimal solution is, for example, the payout amount.
  • the output unit 13 outputs the determined optimal solution to the user terminal device 20, for example.
  • the output unit 13 may output the reason for determining the optimal solution together with the determined optimal solution.
  • the output unit 13 may output the determined optimal solution to a display device (not shown) connected to the decision support system 10. Further, the output unit 13 may output the determined optimal solution to a server that distributes the optimal solution in predicting the race result.
  • the output unit 13 may output the objective function or the weighting coefficient of the feature amount included in the objective function.
  • the weighting coefficient of the feature included in the objective function is the coefficient of the explanatory variable to which the feature corresponds. Further, the output unit 13 may visualize and output the weighting coefficients of the feature amounts included in the objective function. The output unit 13 visualizes and outputs the weighting coefficients included in the objective function by displaying them on a graph, for example.
  • the output unit 13 When the optimal solution in predicting the race result is a combination of voting tickets to be purchased, the output unit 13 outputs the combination of voting tickets indicated as the optimal solution as well as the expected income and expenditure when purchasing voting tickets with the combination. It's okay.
  • the output unit 13 may output the optimal solution for predicting the race result, as well as images of the runners included in the optimal solution and information regarding the runners.
  • the output unit 13 may output either one of the optimal solution in predicting the race result, an image of the runner included in the optimal solution, and information regarding the runner.
  • FIG. 3 shows an example of the display screen of the optimal solution in predicting the race result.
  • the combination of voting tickets to be purchased when purchasing voting tickets for horse racing is displayed as the optimal solution.
  • the optimal solution for the combination of voting tickets to be purchased for the ninth race to be held at the Tokyo Racecourse is displayed.
  • the optimal combination of voting tickets to purchase is displayed when the budget is 1000 yen.
  • the optimal solution for the combination of voting tickets is displayed using the type of voting ticket, horse number or slot number, and purchase amount.
  • odds for each voting ticket are displayed.
  • the combinations of 600 yen for horse group 3-5, 200 yen for horse group 3-9, and 200 yen for frame group 2-3 are displayed as the optimal combinations of voting tickets.
  • FIG. 4 shows an example of a display screen that further displays the refund amount and income and expenditure in the example of the display screen of FIG. 3.
  • the refund amount in the event that the predicted finish is correct is displayed as the refund amount.
  • the bottom row displays the income and expenditure for each horse placed in the race.
  • the balance display in the example of the display screen in FIG. 4 shows that if the first place is horse number 3, and the second place is horse number 5, the balance will be an additional 720 yen.
  • the balance display in the example of the display screen in Figure 4 shows that if 1st place is run by horse number 3 and 2nd place is run by horse number 9, the balance will be an additional 320 yen. .
  • FIG. 5 shows an example of a display screen in which the reason for determining the optimal solution is further displayed in the example of the display screen of FIG. 3.
  • the reason for determining the optimal solution is displayed in the reason column.
  • a feature amount that has a greater influence on determining the optimal solution than other feature amounts is displayed.
  • a feature quantity that has a larger influence on determining the optimal solution than other feature quantities is, for example, a feature quantity whose weighting coefficient is larger than that of the other feature quantities.
  • pedigree and race distance are displayed as reasons for the decision.
  • the example of the display screen in FIG. 5 shows, for example, that the reason why horse race 3-5 is the optimal solution is the pedigree of the horses entering the race.
  • FIG. 6 shows an example of a display screen that displays information on the horses included in the optimal solution.
  • an image of a horse running in lane 3 and information related to the horse running are displayed as reference information.
  • the output unit 13 may add information on a plurality of runners to the optimization result and output the result. Further, the output unit 13 may output information corresponding to a runner clicked or tapped on the display screen, for example.
  • FIG. 7 shows an example of a display screen that displays the objective function used to determine the optimal solution.
  • the expression of the objective function used to determine the optimal solution or the weighting coefficient of the feature quantity included in the objective function is displayed in the prediction expression column at the bottom.
  • the generation unit 14 When the objective function is generated in the decision-making support system 10, the generation unit 14 generates it by inverse reinforcement learning based on the decision-making history regarding prediction of race results in public competitions.
  • the objective function determines the optimal solution for predicting race results from information regarding races in public competitions.
  • the generation unit 14 generates the objective function by learning, for example, a history of decision making regarding prediction of race results in publicly managed competitions by inverse reinforcement learning.
  • the generation unit 14 may generate the objective function as a decision-making model including a combination of feature quantities included as explanatory variables in the objective function, weighting coefficients for each feature quantity, and constraint conditions.
  • the generation unit 14 learns the behavioral data included in the decision-making history regarding prediction of race results in publicly managed competitions as the optimal solution, and generates an optimization index, that is, an objective function.
  • the generation unit 14 generates an objective function by generating a plurality of linear forms and learning rules for selecting the linear forms.
  • the generation unit 14 generates an objective function that reflects the intention corresponding to the decision history by repeatedly performing optimization using a reward and updating the weighting coefficient of the objective function using the decision history.
  • the generation unit 14 When generating the objective function, the generation unit 14 sets the weighting coefficients of the feature amounts included in the objective function using temporary values. The generation unit 14 uses the objective function when performing inverse reinforcement learning using the decision history. The tentative values of the weighting coefficients are set randomly, for example. Then, the generation unit 14 generates an objective function using the set weighting coefficients. After generating the objective function, the generation unit 14 solves the optimization problem using the state data of the decision history and the objective function, and determines an optimal solution. After determining the optimal solution, the generation unit 14 compares the determined optimal solution with behavioral data of the decision-making history.
  • the generation unit 14 updates the weighting coefficients based on the comparison results so that the difference between the optimal solution determined using a predetermined algorithm and the behavioral data of the decision history becomes small, and generates an objective function.
  • the generation unit 14 solves the optimization problem using the state data of the decision history, the generated objective function, and the determined optimal solution, and repeats the process of determining the optimal solution.
  • the generation unit 14 ends the learning when the difference between the determined optimal solution and the behavioral data of the decision history satisfies a criterion set in advance as an end condition. Then, the generation unit 14 stores the generated objective function in the storage unit 15, for example.
  • the generation unit 14 generates the objective function by, for example, maximum entropy inverse reinforcement learning.
  • the generation unit 14 may generate the objective function by sparse inverse reinforcement learning.
  • the generation unit 14 When generating an objective function by sparse inverse reinforcement learning, the generation unit 14 generates the objective function by, for example, assigning weights to all candidates for feature quantities in advance and selecting feature quantities according to the learning results. generate. Further, the generation unit 14 may allocate a plurality of objective functions so that they can be selected according to attributes.
  • a method for generating an objective function is described in, for example, International Publication No. 2021/130915.
  • the storage unit 15 stores, for example, an objective function that the determination unit 12 uses to determine the optimal solution.
  • the storage unit 15 stores the plurality of objective functions.
  • the storage unit 15 may store the updated weighting coefficient of the feature quantity.
  • the storage unit 15 may store decision history for past race predictions, which is used to generate the objective function.
  • Each of the above data may be stored in a storage means other than the storage unit 15.
  • the user terminal device 20 acquires the optimal solution for predicting the race result from the decision support system 10. Then, the user terminal device 20 outputs the optimal solution for predicting the race result to, for example, a display device (not shown).
  • the user terminal device 20 When the user selects an objective function to determine the optimal solution, the user terminal device 20 obtains an input for selecting the objective function input by the user's operation, for example. The user terminal device 20 may acquire the name of the objective function input by the user's operation. The user terminal device 20 then outputs the input result regarding the selection of the input objective function to the decision support system 10.
  • a smartphone, a tablet computer, a notebook computer, or a desktop computer is used as the user terminal device 20.
  • the terminal device used for the user terminal device 20 is not limited to the above example.
  • the information management server 30 is, for example, a server that stores or manages information regarding races in publicly managed competitions.
  • the information management server 30 may be a plurality of servers installed according to the content of information regarding races in public competitions. Information regarding races in public competitions may be stored in a storage device managed by the information management server 30.
  • FIG. 8 is a diagram showing an example of an operation flow when the decision support system 10 determines the optimal solution in predicting the race result.
  • the acquisition unit 11 acquires information regarding races in publicly managed competitions (step S11).
  • the acquisition unit 11 acquires information regarding races in publicly managed competitions, for example, from the information management server 30.
  • the determining unit 12 uses the objective function generated in advance and the acquired information regarding the race in the publicly managed competition to determine the optimal solution for predicting the race result (step S12).
  • the objective function is generated using inverse reinforcement learning based on the decision-making history related to predicting race results in public competitions.
  • the determining unit 12 determines the optimal solution for predicting the race result by solving an optimization problem using, for example, the objective function and the acquired information regarding the publicly managed race.
  • the output unit 13 outputs the optimal solution determined by the determining unit 12 (step S13).
  • the output unit 13 outputs the optimal solution for predicting the race result to the user terminal device 20, for example.
  • the user terminal device 20 After acquiring the optimal solution for predicting the race result, the user terminal device 20 outputs the optimal solution for predicting the race result, for example, to a display device (not shown).
  • FIG. 9 is a diagram illustrating an example of an operation flow when the decision support system 10 generates an objective function.
  • the decision support system 10 uses the generated objective function when determining the optimal solution for predicting race results in public competitions.
  • the acquisition unit 11 acquires a history of decision-making regarding prediction of race results in publicly managed competitions in races held in the past (step S21).
  • the generation unit 14 sets weighting coefficients of the feature amounts included in the objective function using temporary values.
  • the generation unit 14 uses the objective function when performing inverse reinforcement learning using the decision history. After setting the weighting coefficient using the temporary value, the generation unit 14 generates an objective function using the set weighting coefficient (step S22).
  • the generation unit 14 After generating the objective function, the generation unit 14 solves the optimization problem using the decision history and the objective function, and determines the optimal solution (step S23). After determining the optimal solution, the generation unit 14 compares the determined optimal solution with behavioral data of the decision-making history (step S24).
  • the generation unit 14 stores the generated objective function (step S26).
  • the generation unit 14 stores the generated objective function in the storage unit 15, for example.
  • the stored objective function is used by the determining unit 12 to determine the optimal solution.
  • the generation unit 14 updates the weighting coefficient of the feature amount using a preset algorithm (step S27). After updating the weighting coefficients of the feature amounts, the generation unit 14 generates an objective function using the updated weighting coefficients (step S28). After generating the objective function, the generation unit 14 repeats the processing from step S23.
  • the decision-making support system 10 of this embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in publicly managed competitions, and information regarding races in publicly managed competitions. Determine the optimal solution in prediction. Therefore, by using the decision-making support system 10, it becomes possible to easily obtain an optimal solution for predicting the race result based on the decision-making history. Therefore, by using the decision support system 10, it is possible to easily make decisions regarding the race results of publicly managed competitions.
  • the decision support system 10 can perform optimization by, for example, outputting information on the income and expenditure for each combination of voting tickets based on the determination of the optimal solution. This makes it easier for the user to decide whether to apply the result of the decision.
  • the decision support system 10 uses the optimal solution in predicting the race result by using an objective function depending on the person who uses the optimal solution in predicting the race result, for example. It is possible to output the optimal solution suitable for the person who uses the software.
  • FIG. 10 is a diagram showing an example of the information processing system of this embodiment.
  • the information processing system includes a decision support system 40, a user terminal device 20, and an information management server 30.
  • the decision support system 40 is connected to the user terminal device 20 via a network. Further, the decision support system 40 is connected to the information management server 30 via a network.
  • the numbers of user terminal devices 20 and information management servers 30 are set as appropriate.
  • the functions of the user terminal device 20 and the information management server 30 of this embodiment are the same as those of the user terminal device 20 and the information management server 30 of the first embodiment.
  • the decision-making support system 10 of the first embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in publicly managed competitions, and acquired information regarding races in publicly managed competitions. , determine the optimal solution in predicting race results. The decision support system 10 then outputs the optimal solution for predicting the race result of the publicly managed competition.
  • the decision support system 40 of the present embodiment can, for example, change the weighting coefficient of the feature amount included in the objective function used to determine the optimal solution in predicting the race result.
  • the decision support system 40 visualizes and outputs embedded expressions of words that correspond to combinations of weighting coefficients of feature quantities included in the objective function.
  • the words indicate, for example, trends in decision-making regarding prediction of race results in public competitions.
  • a word corresponds to an intention, for example.
  • the decision support system 40 searches for a sentence that includes an embedded expression of a word that corresponds to a combination of weighting coefficients of feature quantities included in the objective function, for example.
  • the decision support system 40 updates the objective function using, for example, a combination of weighting coefficients of features included in the objective function corresponding to the embedded representation of the word.
  • FIG. 11 is a diagram showing an example of the configuration of the decision support system 40.
  • the decision support system 40 includes an acquisition unit 11, a determination unit 12, an output unit 41, a generation unit 14, a storage unit 15, a reception unit 42, an update unit 43, a coefficient prediction unit 44, and an expression prediction unit. 45, a query generation section 46, and a search section 47.
  • the configuration and functions of the acquisition unit 11, determination unit 12, generation unit 14, and storage unit 15 of the decision support system 40 are the same as the acquisition unit 11, determination unit 12, and output unit 13 of the decision support system 10 of the first embodiment. , the generation unit 14 and the storage unit 15, respectively.
  • the output unit 41 outputs, for example, a screen related to changing the weighting coefficient of the objective function. Further, the output unit 41 outputs, for example, a search result obtained by performing a search using a search query.
  • the output unit 41 outputs, for example, a display screen having an input field for the weighting coefficient of the objective function as a screen related to changing the weighting coefficient of the objective function.
  • the output unit 41 may display the weighting coefficients of the objective function as a graph and output a display screen for changing the weighting coefficients on the graph.
  • the output unit 41 When outputting the weighting coefficients of the objective function as a graph, the output unit 41 outputs, for example, a graph centered on feature quantities that make a large contribution to determining the optimal solution by the objective function among the feature quantities.
  • the output unit 41 outputs, for example, a two-dimensional graph centered on two feature quantities that make a large contribution to determining the optimal solution by the objective function among the feature quantities. Further, the output unit 41 may output, for example, a graph centered on feature quantities that have contrasting effects on the determination of the optimal solution by the objective function among the feature quantities.
  • the output unit 41 may output a three-dimensional or more-dimensional graph centered on three or more feature quantities that greatly contribute to determining the optimal solution by the objective function among the feature quantities. Further, the output unit 41 may display the result of converting the multidimensional feature amount into two dimensions through dimension reduction processing as a graph.
  • the output unit 41 outputs, for example, the result of searching for a text using a search query.
  • the text to be searched is, for example, a text related to predicting race results in publicly managed competitions.
  • the sentence to be searched is, for example, a sentence such as ⁇ I started racing in Tokyo 9 races and won big.'' ”
  • the output unit 41 outputs, to the user terminal device 20, for example, a sentence including an expression corresponding to the search query searched by the search unit 47.
  • the output unit 41 outputs posts that include expressions corresponding to the search query searched by the search unit 47.
  • the receiving unit 42 receives, for example, a change value for changing the weighting coefficient of the feature quantity included in the objective function.
  • the receiving unit 42 receives, for example, a change value for changing the weighting coefficient of the feature included in the objective function from the user terminal device 20.
  • a change value for changing the weighting coefficient of the feature quantity included in the objective function is input into the user terminal device 20 by the user of the user terminal device 20, for example.
  • the reception unit 42 may receive the change value from an input device (not shown) connected to the decision support system 40.
  • the reception unit 42 may accept changed values input by changing the weighting coefficients of the feature quantities displayed on the graph. .
  • the receiving unit 42 may receive a change value input by changing the display position of the weighting coefficient of the feature amount on the graph.
  • the reception unit 42 may receive a change value input by changing the display order in a list indicating weighting coefficients of feature amounts.
  • the receiving unit 42 When updating the objective function using a combination of weighting coefficients of feature quantities included in the objective function corresponding to the embedded representation of a word, the receiving unit 42 accepts the combination of weighting coefficients of feature quantities selected by the user. .
  • the reception unit 42 may receive an intention corresponding to a combination of weighting coefficients of feature amounts selected by the user.
  • the reception unit 42 receives the intention by acquiring, from the user terminal device 20, the selection result of the intention, which is input into the user terminal device 20 by the user's operation, for example.
  • the updating unit 43 updates the weighting coefficient of the feature quantity included in the objective function using the obtained value.
  • the updating unit 43 stores, for example, the objective function with updated weighting coefficients in the storage unit 15.
  • the updating unit 43 When updating the objective function based on the accepted intentions, the updating unit 43 extracts, for example, an objective function corresponding to each intention accepted by the accepting unit 42. Then, the updating unit 43 calculates the average value of the weighting coefficients for each feature quantity from the weighting coefficients of the feature quantities included in the extracted objective function. The updating unit 43 updates the objective function using the average value of each of the calculated weighting coefficients as a weighting coefficient of each feature amount. When calculating a new weighting coefficient from the weighting coefficients of the feature amounts included in the plurality of objective functions, the updating unit 43 may calculate the weighting coefficient using a value other than the average value.
  • the updating unit 43 may calculate a new weighting coefficient so that the influence of the weighting coefficient of the feature amount included in the objective function corresponding to the intention to be emphasized becomes greater.
  • the updated objective function is used by the determining unit 12 to determine the optimal solution for predicting the race result.
  • the updating unit 43 may update the objective function using a combination of weighting coefficients of features corresponding to the embedded expression of the word indicating the intention.
  • the updating unit 43 updates the objective function using, for example, a combination of weighting coefficients of feature amounts corresponding to the embedded expression of the word indicating the intention, which is predicted by the coefficient predicting unit 44.
  • FIG. 12 shows an example of a display screen that displays the objective function equation and accepts input for changing the weighting coefficient on the display screen of the optimal solution in predicting the race result in FIG. 3.
  • the objective function equation is displayed in the prediction equation column.
  • the weighting coefficient of the objective function is changed by, for example, rewriting the expression of the weighting coefficient of the objective function by a user's operation and pressing an update button.
  • FIG. 13 shows an example of a display screen that displays weighting coefficients of feature quantities included in the objective function on a two-dimensional graph.
  • the example display screen in FIG. 13 shows the weighting coefficients of x n and x m among the feature amounts included in the objective function as a graph.
  • "Current intention” in the example display screen of FIG. 13 indicates the weighting coefficient of the feature quantity included in the objective function used to determine the optimal solution.
  • “Intention of Mr. A's model” in the example of the display screen in FIG. 13 indicates the weighting coefficient of the feature amount included in the objective function generated using Mr. A's decision-making history.
  • “intention of Mr. B model” in the example of the display screen in FIG. 13 indicates the weighting coefficient of the feature quantity included in the objective function generated using Mr. B's decision-making history.
  • the weighting coefficient may be changed to correspond to the destination position.
  • FIG. 14 shows an example of a display screen that displays feature quantities that make a large contribution to determining the optimal solution in predicting race results as a list in order of magnitude of influence.
  • the value of the weighting coefficient is changed by, for example, changing the arrangement order of the feature amounts by operating a mouse and pressing an update button.
  • the updating unit 43 changes the weighting coefficients of the two feature quantities by exchanging the values of the weighting coefficients of the two feature quantities.
  • FIG. 15 is an example of a display screen that shows weighting coefficients of feature amounts corresponding to two intentions as a graph.
  • weighting coefficients of features included in different objective functions are shown at the upper end and at the latter stage.
  • the weighting coefficient of the feature can be changed by dragging the height of the feature graph. Changed to the value corresponding to the height.
  • the updating unit 43 changes the weighting coefficient of the feature amount using, for example, a value that corresponds to the height of the changed graph.
  • the coefficient prediction unit 44 predicts, for example, a combination of weighting coefficients of feature amounts corresponding to the embedded representation of a word.
  • the coefficient prediction unit 44 predicts a combination of weighting coefficients corresponding to an embedded expression based on a coefficient prediction model that has learned the relationship between an embedded expression of a word and a combination of weighting coefficients of a feature amount.
  • the coefficient prediction model is, for example, a trained model that has learned the relationship between the embedded representation of a word converted into a vector using a natural language processing model and the combination of weighting coefficients of feature amounts.
  • the natural language processing model for example, BERT (Bidirectional Encoder Representations from Transformers) can be used.
  • the expression prediction unit 45 predicts an expression corresponding to the weighting coefficient of the feature amount included in the objective function.
  • the expression prediction unit 45 predicts, for example, an embedded expression of a word corresponding to a weighting coefficient of a feature included in the objective function.
  • the expression prediction unit 45 predicts the embedded expression corresponding to the combination of weighting coefficients based on the expression prediction model that has learned the combination of weighting coefficients and the embedded expression of the word.
  • the embedded representation corresponding to the combination of weighting coefficients includes representations similar to the embedded representation corresponding to the combination of weighting coefficients.
  • FIG. 16 is a diagram schematically showing the correspondence between intention space and language space.
  • the intention space is a space represented by the feature amounts included in the objective function.
  • Linguistic space is a space represented by embedded representations of words.
  • a linguistic space is a space in which the semantic similarity of languages is defined as a continuous value.
  • the vertical and horizontal axes of the graph of the intention space are set using, for example, feature quantities selected from the weighting coefficients of the feature quantities included in the objective function.
  • the vertical and horizontal axes of the language space graph are set based on linguistic meaning. Expressions that are close to each other in the graph of language space are expressions that have similar meanings.
  • intention A in the intention space corresponds to "gamer” in the language space.
  • intention B in the intention space corresponds to "contrarian” in the language space.
  • the expression prediction unit 45 uses a prediction model generated by machine learning using these combinations as training data to predict the embedded expression of the word corresponding to the weighting coefficient of the feature amount.
  • FIG. 17 is a diagram schematically showing the correspondence between the intention space and the language space during prediction using the expression prediction model.
  • intention C in the intention space corresponds to "yamakke" in the language space.
  • the expression prediction model used by the expression prediction unit 45 predicts "Yamake" in the language space for the combination of feature amounts of intention C.
  • FIG. 18 is a diagram schematically showing the correspondence between the intention space and the language space during prediction using the expression prediction model, including the range of similarity in the language space.
  • intention C in the intention space corresponds to "Yamakke” in the language space.
  • Game Master is included in the similar range of "Yamakke”.
  • the expression prediction model may predict "Yamakke” and "Gamer” in the linguistic space for the combination of feature amounts of intention C.
  • FIG. 19 is a diagram schematically showing the correspondence between the language space and the intention space when calculating corresponding weighting coefficients from a plurality of embedded expressions in the language space.
  • the weighting coefficient of the feature amount is calculated using the formula: (c ⁇ (c+b)) ⁇ (weighting factor of intention C)+(b ⁇ (c+b)) ⁇ (weighting factor of intention B).
  • the query generation unit 46 generates a search query based on the embedded expression predicted by the expression prediction unit 45, for example. For example, the query generation unit 46 generates a search query for searching for an embedded expression of a word.
  • the embedded expression of a word corresponds to a combination of weighting coefficients predicted by the expression prediction unit 45 using an expression prediction model.
  • the query generation unit 46 generates, as a search query, a program using SQL to search for an embedded expression of a word corresponding to a combination of weighting coefficients of feature amounts.
  • the search query may be generated using a language other than SQL.
  • the search unit 47 searches for sentences that include expressions that are the same as or similar to the embedded expression of the word, based on the search query. For example, the search unit 47 uses the search query generated by the query generation unit 46 from data on the network to acquire a sentence that includes an expression corresponding to the search query.
  • the search unit 47 obtains posts that include expressions corresponding to the search query from, for example, SNS (Social Networking Service) data.
  • the search unit 47 may, for example, acquire a sentence that includes an expression corresponding to the search query from data on a network other than SNS.
  • FIG. 20 shows an example of a display screen that displays search results for expressions corresponding to the weighting coefficients of the feature quantities included in the objective function as a word cloud.
  • the example display screen in FIG. 20 displays expressions corresponding to combinations of weighting coefficients of feature amounts and similar expressions as a word cloud.
  • the number of occurrences of expressions similar to expressions corresponding to the combination of weighting coefficients of feature amounts in the search results is displayed using a word cloud method.
  • FIG. 21 shows an example of a display screen of search results of SNS posts that include expressions corresponding to weighting coefficients of feature quantities included in the objective function.
  • an SNS post that includes an expression corresponding to the weighting coefficient of the feature amount is displayed.
  • Mr. A is the candidate for the person who is close to the intention of the objective function currently being used.
  • the search unit 47 extracts, for example, Mr. A's posts on the SNS. Then, the output unit 41 outputs Mr. A's post to the user terminal device 20.
  • FIG. 22 shows an example of a display screen for selecting an intention to use for updating the objective function in the language space.
  • the update unit 43 uses the coefficient prediction model to update the weighting coefficient of the feature amount corresponding to the selected intention. Predict the combination of Then, the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction model.
  • FIG. 23 shows an example of a display screen where the intention used to update the objective function is input as text.
  • a sentence containing the intention is input into the input field on the left side.
  • the update unit 43 uses the coefficient prediction model to extract an expression corresponding to the intention from the sentence. A combination of weighting coefficients of features corresponding to the extracted expressions is predicted. Then, the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction model.
  • FIG. 24 is a diagram illustrating an example of an operation flow when the decision support system 40 updates the objective function.
  • the output unit 41 outputs the weighting coefficient of the feature quantity included in the objective function to be updated (step S41).
  • the output unit 41 outputs, to the user terminal device 20, for example, the weighting coefficient of the feature amount included in the objective function to be updated.
  • the receiving unit 42 receives an input of a change value of a weighting coefficient of a feature included in the objective function to be updated (step S42). For example, the receiving unit 42 acquires from the user terminal device 20 a change value of the weighting coefficient of the feature quantity, which is input to the user terminal device 20 by a user's operation.
  • the updating unit 43 updates the objective function using the changed value of the weighting coefficient of the feature quantity acquired by the accepting unit 42 (step S43). After updating the objective function, the updating unit 43 saves the updated objective function (step S44). The updating unit 43 stores the updated objective function in the storage unit 15, for example. The updated objective function is used by the determining unit 12 to determine the optimal solution.
  • FIG. 25 is a diagram illustrating an example of an operation flow when the decision support system 40 updates the objective function based on the embedded expression of words.
  • the output unit 41 outputs the embedded representation of the word corresponding to the weighting coefficient, which is used to update the objective function (step S51).
  • the output unit 41 outputs, for example, the embedded representation of the word corresponding to the weighting coefficient to the user terminal device 20.
  • the accepting unit 42 accepts input of the modified embedded expression (step S52).
  • the reception unit 42 acquires from the user terminal device 20 a modified embedded expression that is input to the user terminal device 20 by a user's operation.
  • the coefficient prediction unit 44 uses the coefficient prediction model to predict the combination of weighting coefficients of the feature amounts corresponding to the acquired embedded expression (step S53).
  • the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction unit 44 (step S54).
  • the updating unit 43 saves the updated objective function (step S55).
  • the updating unit 43 stores the updated objective function in, for example, the storage unit 15.
  • the updated objective function is used by the determining unit 12 to determine the optimal solution.
  • FIG. 26 is a diagram illustrating an example of an operation flow when the decision support system 40 extracts a sentence that includes an expression that corresponds to the weighting coefficient of the feature included in the objective function.
  • the expression prediction unit 45 uses the expression prediction model to predict an expression corresponding to the weighting coefficient of the feature included in the objective function used to determine the optimal solution (step S61).
  • the expression prediction unit 45 uses, for example, an expression prediction model that predicts an expression in language space from the weighting coefficients of the features included in the objective function, and calculates the features included in the objective function used to determine the optimal solution. Predict the embedded representation of the word corresponding to the weighting coefficient.
  • the query generation unit 46 generates a search query to search for sentences that include the corresponding expression (step S62).
  • the search unit 47 uses, for example, the search query generated by the query generation unit 46 to search the data on the network for a sentence corresponding to the search query (step S63).
  • the search unit 47 obtains posts corresponding to the search query from, for example, SNS data on the network.
  • the output unit 41 When a sentence including an expression corresponding to the search query is acquired, the output unit 41 outputs a search result using the search query (step S64). The output unit 41 outputs a search result using a search query to the user terminal device 20, for example.
  • the decision support system 40 obtains the change value of the weighting coefficient of the feature quantity included in the objective function.
  • the decision support system 40 then updates the objective function using the changed value.
  • the updated objective function for example, it is possible to determine an optimal solution that is optimized depending on the person who will use the optimal solution regarding prediction of race results.
  • the decision support system 40 When displaying the weighting coefficients of the feature quantities included in the objective function on a graph and accepting input of change values, the decision support system 40 changes the objective function based on the result of changing the display state of the weighting coefficients on the graph. Update. By updating the objective function in this way, optimization of the objective function becomes easier. By optimizing the objective function, it is possible to obtain an optimal solution that is more suitable for the person who uses the optimal solution in predicting the race result.
  • the user when updating the objective function based on the selection result of the intention corresponding to the objective function, the user can easily update the objective function even if the user does not have knowledge about changing the weighting coefficient of the feature amount.
  • the decision support system 40 extracts sentences that include expressions in the linguistic space that correspond to the weighting coefficients of the features included in the objective function.
  • the decision support system 40 then outputs the extracted text to the user terminal device 20, for example.
  • the decision support system 40 can output information related to the reason for determining the optimal solution.
  • the user of the decision support system 40 can refer to information related to the reason for determining the optimal solution and interpret the result of determining the optimal solution.
  • FIG. 27 is a diagram showing an example of the configuration of the decision support system 100 of this embodiment.
  • the decision support system 100 includes an acquisition section 101, a determination section 102, and an output section 103.
  • the acquisition unit 101 acquires information regarding races in publicly managed competitions.
  • the determining unit 102 uses an objective function generated in advance by inverse reinforcement learning based on the decision history regarding the prediction of race results in publicly managed competitions, and the obtained information regarding races in publicly managed competitions to determine the optimal solution for predicting race results. Determine.
  • the output unit 103 outputs information regarding prediction of the race result based on the determined optimal solution.
  • the acquisition unit 11 of the first embodiment is an example of the acquisition unit 101. Further, the acquisition unit 101 is one aspect of an acquisition means.
  • the determining unit 12 of the first embodiment is an example of the determining unit 102. Furthermore, the determining unit 102 is one aspect of determining means.
  • the output unit 13 of the first embodiment is an example of the output unit 103. Furthermore, the output unit 103 is one aspect of output means.
  • FIG. 28 is a diagram illustrating an example of the operation flow of the decision support system 100.
  • the acquisition unit 101 acquires information regarding races in publicly managed competitions (step S101).
  • the determining unit 102 uses an objective function generated in advance by inverse reinforcement learning based on the decision history regarding prediction of the race result in the publicly managed competition and the acquired information regarding the race in the publicly managed competition.
  • the optimal solution for predicting the race result is determined using the following (step S102). Once the optimal solution is determined, the output unit 103 outputs information regarding prediction of the race result based on the determined optimal solution (step S103).
  • the decision-making support system 100 of this embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding the prediction of race results in publicly managed competitions, and information regarding races in publicly managed competitions. Determine the optimal solution for prediction. The decision support system 100 then outputs the determined optimal solution. As a result, by using the decision-making support system 100, it is possible to easily make decisions regarding the race results of publicly managed competitions.
  • FIG. 29 shows a computer that executes a computer program that performs each process in the decision support system 10 of the first embodiment, the decision support system 40 of the second embodiment, and the decision support system 100 of the third embodiment.
  • 200 shows an example of a configuration of 200.
  • the computer 200 includes a CPU (Central Processing Unit) 201, a memory 202, a storage device 203, an input/output I/F (Interface) 204, and a communication I/F 205.
  • CPU Central Processing Unit
  • the CPU 201 reads computer programs for performing each process from the storage device 203 and executes them.
  • the CPU 201 may be configured by a combination of multiple CPUs. Further, the CPU 201 may be configured by a combination of a CPU and other types of processors. For example, the CPU 201 may be configured by a combination of a CPU and a GPU (Graphics Processing Unit).
  • the memory 202 is configured with a DRAM (Dynamic Random Access Memory) or the like, and temporarily stores computer programs executed by the CPU 201 and data being processed.
  • the storage device 203 stores computer programs executed by the CPU 201.
  • the storage device 203 is configured by, for example, a nonvolatile semiconductor storage device. Other storage devices such as a hard disk drive may be used as the storage device 203.
  • the input/output I/F 204 is an interface that receives input from a worker and outputs display data and the like.
  • the communication I/F 205 is an interface that transmits and receives data between the user terminal device 20 and the information management server 30. Further, the user terminal device 20 and the information management server 30 may also have similar configurations.
  • the computer program used to execute each process can also be stored and distributed in a computer-readable recording medium that non-temporarily records data.
  • a computer-readable recording medium for example, a magnetic tape for data recording or a magnetic disk such as a hard disk can be used.
  • an optical disc such as a CD-ROM (Compact Disc Read Only Memory) can also be used.
  • a nonvolatile semiconductor memory device may be used as the recording medium.
  • the output means outputs data visualizing combinations of weighting coefficients of features included in the objective function.
  • Decision support system according to appendix 1 or 2.
  • reception means for accepting a change value for changing a weighting coefficient of a feature included in the objective function; and updating means for updating the objective function using the received change value,
  • the determining means determines the optimal solution using the updated objective function.
  • the output means outputs weighting coefficients of features included in the objective function as a graph,
  • the receiving means receives the changed value input by changing a weighting coefficient of a feature included in the objective function displayed on the graph. Decision support system described in Appendix 4.
  • Coefficient prediction means for predicting the combination of weighting coefficients from the embedded representation of the word using a coefficient prediction model that predicts the combination of weighting coefficients of the feature quantities included in the objective function from the embedded representation of the word; further comprising: updating means for updating the objective function based on the combination of the weighting coefficients predicted by the coefficient predicting means; The decision support system described in any of Supplementary Notes 1 to 3.
  • a combination of weighting coefficients is calculated from a combination of weighting coefficients of features included in the objective function.
  • Decision Support System 11 Acquisition Unit 12 Determination Unit 13 Output Unit 14 Generation Unit 15 Storage Unit 20 User Terminal Device 30 Information Management Server 40 Decision Support System 41 Output Unit 42 Reception Unit 43 Update Unit 44 Coefficient Prediction Unit 45 Expression Prediction Section 46 Query generation section 47 Search section 100 Decision support system 101 Acquisition section 102 Determination section 103 Output section 200 Computer 201 CPU 202 Memory 203 Storage device 204 Input/output I/F 205 Communication I/F

Landscapes

  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

This decision support system comprises an acquisition unit, a decision unit, and an output unit. The acquisition unit acquires information relating to a public competition race. The decision unit decides the optimal solution for a prediction of race results using acquired information relating to the public competition race, and an objective function generated in advance by inverse reinforcement learning based on a decision history relating to predictions of public competition race results. The output unit outputs the information relating to the prediction of the race results on the basis of the optimal solution that has been decided.

Description

意思決定支援システム、意思決定支援方法および記録媒体Decision support system, decision support method and recording medium
 本発明は、意思決定支援システム等に関する。 The present invention relates to a decision support system and the like.
 公営競技では、例えば、レース条件、およびレースに出場する競技体の状態に関する多くの要素がレース結果に影響を及ぼし得る。多くの要素の影響がレース結果の予測を難しくする一方で、予測の難しさは、レース結果の予想が愛好者の楽しみの1つとなり得る。しかし、例えば、初心者にとっては、レース結果を予測する際に、レースに関する情報のどの項目に着目するかが難しい場合がある。そのため、レース結果の予測に関する情報を提供できるシステムがあることが望ましい。 In public competitions, for example, many factors related to race conditions and the condition of the athletes participating in the race can affect the race result. While many factors make it difficult to predict race results, the difficulty of prediction can be part of the fun of predicting race results for enthusiasts. However, for example, for beginners, it may be difficult to decide which item of information regarding the race to focus on when predicting the race result. Therefore, it is desirable to have a system that can provide information regarding predictions of race results.
 特許文献1の学習装置は、意思決定履歴データに基づく逆強化学習によって、最適解を決定するための目的関数を生成する。 The learning device of Patent Document 1 generates an objective function for determining an optimal solution by inverse reinforcement learning based on decision history data.
 特許文献2の番組編成プログラムは、学習モデルを用いてレースに出場する選手を編成する。 The program programming program in Patent Document 2 uses a learning model to organize athletes who will participate in a race.
国際公開第2021/229626号International Publication No. 2021/229626 特開2020-60875号公報JP2020-60875A
 特許文献1の学習装置および特許文献2の番組編成プログラムは、公営競技のレース結果に関する意思決定において最適な案を提示できない場合がある。 The learning device of Patent Document 1 and the program programming program of Patent Document 2 may not be able to present optimal proposals in decision-making regarding race results of publicly managed competitions.
 上記の課題を解決するため、公営競技のレース結果に関する意思決定を容易に行うことができる意思決定支援システム等を提供することを目的とする。 In order to solve the above problems, the purpose is to provide a decision-making support system, etc. that can easily make decisions regarding the race results of publicly managed competitions.
 上記の課題を解決するため、本発明の意思決定支援システムは、公営競技のレースに関する情報を取得する取得手段と、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する決定手段と、決定した最適解を基に、レース結果の予測に関する情報を出力する出力手段とを備える。 In order to solve the above problems, the decision support system of the present invention includes an acquisition means for acquiring information regarding races in publicly managed competitions, and a decision support system that is generated in advance by inverse reinforcement learning based on the decision history regarding prediction of race results in publicly managed competitions. a determining means for determining an optimal solution for predicting race results using an objective function determined by the objective function and acquired information regarding races in public competitions; and an output for outputting information regarding predicting race results based on the determined optimal solution. and means.
 本発明の意思決定支援方法は、公営競技のレースに関する情報を取得し、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定し、決定した最適解を基に、レース結果の予測に関する情報を出力する。 The decision-making support method of the present invention acquires information regarding races in publicly managed competitions, and uses objective functions generated in advance by inverse reinforcement learning based on decision history regarding prediction of race results in publicly managed competitions, and the acquired race results in publicly managed competitions. The optimal solution for predicting the race result is determined using the information regarding the race result, and information regarding the prediction of the race result is output based on the determined optimal solution.
 本発明の記録媒体は、公営競技のレースに関する情報を取得する処理と、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する処理と、決定した最適解を基に、レース結果の予測に関する情報を出力する処理とをコンピュータに実行させる意思決定支援プログラムを非一時的に記録する。 The recording medium of the present invention includes a process for acquiring information regarding races in publicly managed competitions, an objective function generated in advance by inverse reinforcement learning based on a decision history regarding prediction of race results in publicly managed competitions, and an acquired race result in publicly managed competitions. A non-decision support program that causes a computer to execute the process of determining the optimal solution for predicting the race result using information about the race result, and the process of outputting information regarding the prediction of the race result based on the determined optimal solution. Record temporarily.
 本発明によると、公営競技のレース結果に関する意思決定を容易に行うことができる。 According to the present invention, decisions regarding the race results of publicly managed competitions can be easily made.
本発明の第1の実施形態の構成の一例を示す図である。1 is a diagram showing an example of a configuration of a first embodiment of the present invention. FIG. 本発明の第1の実施形態の意思決定支援システムの構成の例を示す図である。1 is a diagram showing an example of the configuration of a decision support system according to a first embodiment of the present invention. 本発明の第1の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 1st embodiment of the present invention. 本発明の第1の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 1st embodiment of the present invention. 本発明の第1の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 1st embodiment of the present invention. 本発明の第1の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 1st embodiment of the present invention. 本発明の第1の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 1st embodiment of the present invention. 本発明の第1の実施形態の意思決定支援システムの動作フローの例を示す図である。FIG. 3 is a diagram showing an example of the operation flow of the decision support system according to the first embodiment of the present invention. 本発明の第1の実施形態の意思決定支援システムの動作フローの例を示す図である。FIG. 3 is a diagram showing an example of the operation flow of the decision support system according to the first embodiment of the present invention. 本発明の第2の実施形態の構成の一例を示す図である。It is a figure showing an example of composition of a 2nd embodiment of the present invention. 本発明の第2の実施形態の意思決定支援システムの構成の例を示す図である。It is a figure showing an example of composition of a decision support system of a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における言語空間と意図空間との対応関係を模式的に示す図である。FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention. 本発明の第2の実施形態における言語空間と意図空間との対応関係を模式的に示す図である。FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention. 本発明の第2の実施形態における言語空間と意図空間との対応関係を模式的に示す図である。FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention. 本発明の第2の実施形態における言語空間と意図空間との対応関係を模式的に示す図である。FIG. 7 is a diagram schematically showing the correspondence between language space and intention space in the second embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態における表示画面の例を示す図である。It is a figure showing an example of a display screen in a 2nd embodiment of the present invention. 本発明の第2の実施形態の意思決定支援システムの動作フローの例を示す図である。It is a figure showing an example of an operation flow of a decision-making support system of a 2nd embodiment of the present invention. 本発明の第2の実施形態の意思決定支援システムの動作フローの例を示す図である。It is a figure showing an example of an operation flow of a decision-making support system of a 2nd embodiment of the present invention. 本発明の第2の実施形態の意思決定支援システムの動作フローの例を示す図である。It is a figure showing an example of an operation flow of a decision-making support system of a 2nd embodiment of the present invention. 本発明の第3の実施形態の意思決定支援システムの構成の例を示す図である。It is a figure showing an example of composition of a decision-making support system of a 3rd embodiment of the present invention. 本発明の第3の実施形態の意思決定支援システムの動作フローの例を示す図である。It is a figure showing an example of an operation flow of a decision-making support system of a 3rd embodiment of the present invention. 本発明の他の実施形態の構成の例を示す図である。It is a figure showing the example of composition of other embodiments of the present invention.
 本発明の第1の実施形態について、図を参照して詳細に説明する。図1は、本実施形態の情報処理システムの例を示す図である。一例として、情報処理システムは、意思決定支援システム10と、利用者端末装置20と、情報管理サーバ30を備える。意思決定支援システム10は、ネットワークを介して、利用者端末装置20と接続する。また、意思決定支援システム10は、ネットワークを介して、情報管理サーバ30と接続する。 A first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an example of an information processing system according to this embodiment. As an example, the information processing system includes a decision support system 10, a user terminal device 20, and an information management server 30. The decision support system 10 is connected to a user terminal device 20 via a network. Further, the decision support system 10 is connected to an information management server 30 via a network.
 意思決定支援システム10は、公営競技のレース結果に関する意思決定を支援するシステムである。公営競技は、例えば、競馬である。公営競技は、競輪、競艇またはオートレースであってもよい。公営競技の例は、上記に限られず、公的機関がギャンブルとして開催する競技であれば、競技の種類は問わない。 The decision support system 10 is a system that supports decision making regarding the race results of publicly managed competitions. An example of a publicly managed game is horse racing. Publicly managed competitions may be bicycle races, boat races, or auto races. Examples of publicly managed competitions are not limited to those mentioned above, and the type of competition does not matter as long as it is a competition held by a public institution for gambling purposes.
 公営競技のレース結果に関する意思決定は、例えば、公営競技のレースの投票券を買う際の、投票券の組み合わせを決定することである。公営競技のレース結果に関する意思決定は、レースの着順予測の決定、または投票券を購入するレースの決定であってもよい。公営競技が競馬である場合に、投票券の組み合わせの決定は、例えば、投票券の購入において、馬番号二連勝複式(馬連)と、枠番号二連勝式複式(枠連)とをどのような組み合わせで買うかを決定することである。投票券の組み合わせは、上記の例に限られない。また、公営競技のレース結果に関する意思決定の例は、投票券の組み合わせの決定に限られない。 Decision-making regarding race results in publicly managed competitions is, for example, determining the combination of voting tickets when purchasing voting tickets for races in publicly managed competitions. The decision regarding the race result of the publicly managed competition may be a decision on predicting the finish order of the race or a decision on the race for which voting tickets are to be purchased. When the publicly managed competition is horse racing, the combination of voting tickets is determined by determining, for example, when purchasing voting tickets, what kind of horse numbers are used for two consecutive wins (horse racing) and frame numbers are two consecutive winning combinations (fraction). The key is to decide whether to buy a combination. The combination of voting tickets is not limited to the above example. Furthermore, examples of decision-making regarding race results in publicly managed competitions are not limited to determining the combination of voting tickets.
 意思決定支援システム10は、例えば、予め生成された目的関数を用いて、レース結果の予測における最適解を決定する。目的関数は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習によって生成される。レース結果の予測における最適解は、最適化結果ともいう。そして、意思決定支援システム10は、利用者端末装置20に、決定した最適解を基に、レース結果に関する情報を出力する。目的関数は、レース結果の予測における最適解を決定する際に用いられる関数である。特徴量は、目的関数において、説明変数が対応する。特徴量は、例えば、意思決定に影響を与え得るレースに関する情報である。公営競技が競馬である場合に、特徴量は、例えば、血統およびレースの距離である。特徴量は、血統およびレースの距離に限られない。また、特徴量の重み係数は、目的関数において、説明変数の係数が対応する。特徴量の重み係数は、意思決定においてレースに関する情報をどれだけ重視するかを示す指標である。 The decision support system 10 uses, for example, a pre-generated objective function to determine the optimal solution for predicting race results. The objective function is generated by inverse reinforcement learning based on the history of decision-making regarding prediction of race results in public competitions. The optimal solution in predicting race results is also called an optimization result. The decision support system 10 then outputs information regarding the race results to the user terminal device 20 based on the determined optimal solution. The objective function is a function used to determine the optimal solution for predicting race results. The feature amounts correspond to explanatory variables in the objective function. The feature amount is, for example, information regarding a race that can influence decision making. When the public competition is horse racing, the feature amounts are, for example, pedigree and race distance. Features are not limited to pedigree and race distance. Further, the weighting coefficient of the feature amount corresponds to the coefficient of the explanatory variable in the objective function. The weighting coefficient of the feature amount is an index indicating how much importance is given to information regarding the race in decision making.
 公営競技のレース結果の予測に関する意思決定履歴に基づいて生成された目的関数に含まれる特徴量と、特徴量の重み係数の組み合わせは、意図とも呼ばれる。目的関数に含まれる特徴量と、特徴量の重み係数の組み合わせは、目的関数を生成する際に意思決定履歴が用いられた人物の意思決定に関する意図を反映している。意図は、例えば、意思決定の傾向に応じた自然言語(単語、文章を含む)を用いて説明される。例えば、目的関数における所定値以上の重み係数の特徴量を、意思決定の傾向に応じた単語としてもよい。具体的には、当該特徴量が脚質である場合、「脚質」を意思決定の傾向に応じた単語とする。目的関数は、例えば、熟練者の意図を反映して生成される。熟練者が複数の場合に、目的関数は、例えば、熟練者それぞれの意図を反映して、複数、生成されてもよい。意思決定支援システム10は、意図ごとに生成された複数の目的関数のうち、いずれかの目的関数を用いてレース結果の予測における最適解を決定してもよい。 The combination of the feature quantities included in the objective function generated based on the decision-making history regarding prediction of race results in public competitions and the weighting coefficients of the feature quantities is also called intention. The combination of the feature amounts included in the objective function and the weighting coefficients of the feature amounts reflects the decision-making intention of the person whose decision history is used when generating the objective function. The intention is explained using natural language (including words and sentences) according to decision-making tendencies, for example. For example, a feature amount having a weighting coefficient greater than or equal to a predetermined value in the objective function may be used as a word corresponding to a decision-making tendency. Specifically, when the feature amount is leg quality, "leg quality" is set as a word according to the tendency of decision making. The objective function is generated, for example, reflecting the intention of an expert. When there are multiple experts, a plurality of objective functions may be generated, for example, reflecting the intentions of each expert. The decision support system 10 may determine the optimal solution for predicting the race result using any one of the plurality of objective functions generated for each intention.
 公営競技のレース結果の予測に関する意思決定履歴は、公営競技に関する状態データと、公営競技に関する行動データを含む。公営競技に関する状態データは、例えば、公営競技のレースに関する情報である。公営競技のレースに関する情報は、例えば、レースの条件、オッズおよびレースに出場する競技体の属性である。競技体は、レースに参加する主体である。競技体の属性は、レースに参加する競技体それぞれの情報である。公営競技が競馬である場合に、競技体は、馬である。公営競技が競馬である場合に、競技体の属性には、騎手に関する情報も含まれる。公営競技が競輪、競艇またはオートレースである場合に、競技体は、選手である。競技体の属性には、自転車、ボートまたはオートバイの情報が含まれていてもよい。レースに関する情報は、上記の例に限られない。 The decision-making history regarding the prediction of the race result of a publicly managed competition includes status data regarding the publicly managed competition and behavioral data regarding the publicly managed competition. The status data regarding publicly managed competitions is, for example, information regarding races in publicly managed competitions. Information regarding races in public competitions includes, for example, race conditions, odds, and attributes of athletes participating in the races. The competition body is the main body that participates in the race. The attributes of the competition objects are information about each competition object participating in the race. When the publicly managed competition is horse racing, the competition object is a horse. When the publicly managed sport is horse racing, the attributes of the sport also include information regarding the jockey. When the publicly managed competition is a bicycle race, a boat race, or an auto race, the competing objects are athletes. The attributes of the competition object may include information about a bicycle, a boat, or a motorcycle. Information regarding races is not limited to the above example.
 公営競技に関する行動データは、例えば、公営競技のレースに関してある人物が意図を持って行う意思決定の結果である。意図を持って行う意思決定は、例えば、意思決定を行う人物が、当該人物が有している方針に基づいて行う意思決定である。例えば、意思決定が投票券を買う際の組み合わせに関する決定である場合に、意思決定履歴は、例えば、過去のあるレースに関する情報と、当該レースに対して熟練者が行った投票券の組み合わせに関する意思決定である。投票券の組み合わせに関する意思決定は、例えば、投票券の購入パターンの決定である。投票券の購入パターンは、例えば、購入する投票券の組み合わせである。また、投票券の購入パターンは、投票券の組み合わせと、投票券ごとの購入額であってもよい。この場合、過去のあるレースに関する情報は、状態データである。また、レースに対して熟練者が意思決定した投票券の組み合わせは、行動データである。また、意思決定履歴は、専門家、または解説者の意思決定のデータであってもよい。 Behavioral data related to publicly managed competitions is, for example, the result of a decision made intentionally by a person regarding a race in a publicly managed competition. A decision made with intention is, for example, a decision made by a person making the decision based on a policy that the person has. For example, when a decision is about a combination of voting tickets to purchase, the decision history includes information about a certain race in the past and the decision made by an expert regarding the combination of voting tickets for that race. It's a decision. The decision regarding the combination of voting tickets is, for example, the determination of the purchasing pattern of voting tickets. The voting ticket purchase pattern is, for example, a combination of voting tickets to be purchased. Moreover, the voting ticket purchase pattern may be a combination of voting tickets and a purchase amount for each voting ticket. In this case, the information regarding a certain race in the past is status data. Further, the combination of voting tickets determined by the expert for the race is behavioral data. Further, the decision-making history may be data on decisions made by experts or commentators.
 意思決定支援システム10は、例えば、過去のレースにおける、公営競技のレース結果の予測に関する意思決定履歴に基づき、逆強化学習によって目的関数を生成する。生成された目的関数は、レース結果の予測における最適解を決定するために用いられる。意思決定支援システム10は、例えば、過去のレースにおける、レースに関する情報と、意思決定履歴を用いた逆強化学習によって、目的関数に含まれる特徴量の重み係数を決定することで目的関数を生成する。目的関数に含まれる特徴量の重み係数は、目的関数において特徴量に対応する説明変数の係数である。意思決定支援システム10は、意思決定支援システム10の外部で生成された目的関数を用いて、レース結果の予測における最適解を決定してもよい。 The decision-making support system 10 generates an objective function by inverse reinforcement learning, for example, based on the decision-making history regarding prediction of race results in publicly managed competitions in past races. The generated objective function is used to determine the optimal solution in predicting race results. The decision support system 10 generates an objective function by determining weighting coefficients of features included in the objective function, for example, by inverse reinforcement learning using race information and decision history in past races. . The weighting coefficient of the feature included in the objective function is the coefficient of the explanatory variable corresponding to the feature in the objective function. The decision support system 10 may use an objective function generated outside the decision support system 10 to determine the optimal solution for predicting the race result.
 利用者端末装置20は、例えば、意思決定支援システム10が決定するレース結果の予測における最適解を利用する人物が所持している端末装置である。レース結果の予測における最適解を利用する人物は、例えば、レースの投票券を購入する人物である。レース結果の予測における最適解を利用する人物は、レースの編成を考案する担当者であってもよい。また、レース結果の予測における最適解を利用する人物は、記者または解説者であってもよい。レース結果の予測における最適解を利用する人物は、上記の例に限られない。 The user terminal device 20 is, for example, a terminal device owned by a person who uses the optimal solution in predicting the race result determined by the decision support system 10. A person who uses the optimal solution in predicting a race result is, for example, a person who purchases a voting ticket for a race. The person who utilizes the optimal solution in predicting the race result may be the person in charge of devising the organization of the race. Further, the person who uses the optimal solution in predicting the race result may be a reporter or a commentator. The person who uses the optimal solution in predicting the race result is not limited to the above example.
 情報管理サーバ30は、例えば、公営競技のレースに関する情報を保有しているサーバである。レースに関する情報は、例えば、レースの条件およびレースに出場する競技体の属性である。競技体は、レースに参加する主体である。競技体の属性は、レースに参加する競技体それぞれの情報である。公営競技が競馬である場合に、競技体は、馬である。公営競技が競馬である場合に、競技体の属性には、騎手に関する情報も含まれる。公営競技が競輪、競艇またはオートレースである場合に、競技体は、選手である。競技体の属性には、自転車、ボートまたはオートバイの情報が含まれていてもよい。レースに関する情報は、上記の例に限られない。 The information management server 30 is, for example, a server that holds information regarding races in publicly managed competitions. The information regarding the race is, for example, the conditions of the race and the attributes of the athletes participating in the race. The competition body is the main body that participates in the race. The attributes of the competition objects are information about each competition object participating in the race. When the publicly managed competition is horse racing, the competition object is a horse. When the publicly managed sport is horse racing, the attributes of the sport also include information regarding the jockey. When the publicly managed competition is a bicycle race, a boat race, or an auto race, the competing objects are athletes. The attributes of the competition object may include information about a bicycle, a boat, or a motorcycle. Information regarding races is not limited to the above example.
 意思決定支援システム10は、例えば、情報管理サーバ30から公営競技のレースに関する情報を取得する。そして、意思決定支援システム10は、情報管理サーバ30から取得した公営競技のレースに関する情報と、目的関数を用いてレース結果の予測における最適解を決定する。そして、意思決定支援システム10は、利用者端末装置20に、レース結果の予測における最適解を出力する。 For example, the decision support system 10 acquires information regarding races in public competitions from the information management server 30. Then, the decision support system 10 determines the optimal solution for predicting the race result using the information regarding the race of the publicly managed competition acquired from the information management server 30 and the objective function. The decision support system 10 then outputs the optimal solution for predicting the race result to the user terminal device 20.
 意思決定支援システム10は、複数の情報管理サーバ30から公営競技のレースに関する情報を取得してもよい。また、意思決定支援システム10は、利用者端末装置20から、公営競技のレースに関する情報を取得してもよい。 The decision support system 10 may acquire information regarding races in public competitions from a plurality of information management servers 30. Further, the decision support system 10 may acquire information regarding races in public competitions from the user terminal device 20.
 意思決定支援システム10は、複数の利用者端末装置20に、レース結果の予測における最適解を出力してもよい。意思決定支援システム10は、例えば、複数の利用者がそれぞれ利用している利用者端末装置20に、レース結果の予測における最適解を出力してもよい。利用者端末装置20および情報管理サーバ30の数は、適宜、設定され得る。 The decision support system 10 may output the optimal solution for predicting the race result to the plurality of user terminal devices 20. For example, the decision support system 10 may output the optimal solution for predicting the race result to the user terminal devices 20 used by a plurality of users. The number of user terminal devices 20 and information management servers 30 may be set as appropriate.
 意思決定支援システム10の構成について説明する。図2は、意思決定支援システム10の構成の例を示す図である。意思決定支援システム10は、取得部11と、決定部12と、出力部13と、生成部14と、記憶部15を備える。 The configuration of the decision support system 10 will be explained. FIG. 2 is a diagram showing an example of the configuration of the decision support system 10. The decision support system 10 includes an acquisition section 11 , a determination section 12 , an output section 13 , a generation section 14 , and a storage section 15 .
 取得部11は、公営競技のレースに関する情報を取得する。公営競技のレースに関する情報は、例えば、レース結果の予測における最適解の決定に関係し得る情報である。取得部11は、公営競技のレースに関する情報として、例えば、レース条件と、オッズと、競技体の属性とを取得する。取得部11は、投票券の種類ごとにオッズを取得する。 The acquisition unit 11 acquires information regarding races in publicly managed competitions. Information regarding races in public competitions is, for example, information that can be related to determining the optimal solution in predicting race results. The acquisition unit 11 acquires, for example, race conditions, odds, and attributes of competition objects as information regarding races in publicly managed competitions. The acquisition unit 11 acquires odds for each type of voting ticket.
 公営競技が競馬の場合に、レース条件には、例えば、競技場に関する情報が含まれる。また、レース条件には、レースの設定条件と、出走馬が満たすべき条件が含まれていてもよい。レース条件は、例えば、距離、馬場の種類、出走馬の条件、斤量、レースの格付け、レース場、天候、馬場の状態、出走馬の数および走行方向のうち少なくも1つ以上である。馬場の種類は、例えば、芝、ダートまたは障害の区別である。出走馬の条件は、例えば、馬齢および性別によって規定される出走の条件である。馬場の状態は、例えば、馬場が含む水分量を示す情報である。走行方向は、例えば、レースが左回りで行われるか右回りで行われるかを示す情報である。公営競技が競馬の場合におけるレース条件は、上記の例に限られない。 If the publicly managed competition is horse racing, the race conditions include, for example, information regarding the stadium. Further, the race conditions may include race setting conditions and conditions that the runners must satisfy. The race conditions include, for example, at least one of distance, race track type, race horse conditions, weight, race rating, race track, weather, race track condition, number of race horses, and running direction. The type of horse track is, for example, grass, dirt, or obstacles. The conditions for a horse to run are defined by, for example, the age and sex of the horse. The condition of the horse track is, for example, information indicating the amount of water contained in the horse track. The running direction is, for example, information indicating whether the race is run counterclockwise or clockwise. Race conditions when the publicly managed competition is horse racing are not limited to the above example.
 公営競技が競馬の場合に、競技体の属性は、出走馬の属性である。出走馬の属性は、出走馬それぞれの情報である。出走馬の属性は、例えば、レーン番号、枠番号、オッズ、年齢、性別、体重、体重変化、血液データ、筋肉量、調教状況、健康状態、休養履歴、レース出場履歴、負担斤量、脚質、戦績、父馬、母馬、馬主、厩舎、調教師および生産者のうち少なくも1つ以上である。レーン番号は、馬番であってもよい。調教状況は、例えば、調教時の距離ごとのタイムおよびタイム変化である。脚質は、例えば、逃げ馬、先行馬、差し馬または追い込み馬の区分によって設定される。出走馬の属性には、父馬および母馬の戦績が含まれていてもよい。また、戦績は、例えば、過去のレースにおけるレース条件、レース時における出走馬の属性、獲得賞金およびレース展開である。レース展開は、例えば、位置取りおよび着差である。位置取りは、例えば、レースの全区間を所定の距離ごとに分割した場合における、各区間における順位およびタイムである。着差は、例えば、順位が上の出走馬または順位が下の出走馬とのタイム差である。出走馬の属性は、上記の例に限られない。 If the publicly managed competition is horse racing, the attributes of the competition are the attributes of the horses running. The attributes of the runners are information about each runner. The attributes of the running horses include, for example, lane number, slot number, odds, age, gender, weight, weight change, blood data, muscle mass, training status, health condition, rest history, race participation history, weight load, leg quality, At least one of race record, sire horse, dam horse, owner, stable, trainer, and producer. The lane number may be a horse number. The training situation is, for example, time and time change for each distance during training. The foot quality is set, for example, by classification of a runaway horse, a leading horse, a lead horse, or a chasing horse. The attributes of the running horses may include the race records of the sire horse and dam horse. Furthermore, race results include, for example, race conditions in past races, attributes of horses running in races, prize money earned, and race developments. Lace development includes, for example, positioning and difference in placement. The positioning is, for example, the ranking and time in each section when the entire section of the race is divided into predetermined distances. The difference in finish is, for example, the time difference between a horse that is higher in the ranking or a horse that is lower in the ranking. The attributes of the running horses are not limited to the above example.
 公営競技が競輪の場合に、レースの条件は、例えば、レース場、競争距離および天候のうち少なくも1つ以上である。また、競技体の属性は、選手の身長、体重、年齢、脚質、オッズおよび戦績のうち少なくも1つ以上である。公営競技が競輪の場合における、レースの条件および競技体の属性の例は、上記に限られない。 When the public competition is bicycle racing, the race conditions include, for example, at least one of the race track, race distance, and weather. Further, the attributes of the competition body include at least one of the athlete's height, weight, age, leg quality, odds, and match record. In the case where the publicly managed race is bicycle racing, examples of race conditions and attributes of the race bodies are not limited to the above.
 公営競技が競艇の場合に、レースの条件は、例えば、レース場、競争距離および天候のうち少なくも1つ以上である。また、競技体の属性は、選手の身長、体重、年齢、ランク、オッズおよび戦績のうち少なくも1つ以上である。公営競技が競艇の場合における、レースの条件および競技体の属性の例は、上記に限られない。 When the publicly managed competition is a boat race, the race conditions include, for example, at least one of the race track, race distance, and weather. Further, the attributes of the competition body include at least one of the athlete's height, weight, age, rank, odds, and match record. In the case where the publicly managed competition is a boat race, examples of race conditions and attributes of competition objects are not limited to the above.
 公営競技がオートレースの場合に、レースの条件は、例えば、レース場、ハンデの有無および天候のうち少なくも1つ以上である。また、競技体の属性は、選手の身長、体重、年齢、所属、階級、オッズおよび戦績のうち少なくも1つ以上である。競技体の属性は、スタート展示情報を含んでいてもよい。公営競技がオートレースの場合におけるレースの条件および競技体の属性の例は、上記に限られない。 When the public competition is an auto race, the race conditions include, for example, at least one of the race track, the presence or absence of a handicap, and the weather. Further, the attributes of the athlete are at least one of the athlete's height, weight, age, affiliation, class, odds, and match record. The attributes of the competition object may include start display information. Examples of race conditions and attributes of race objects when the publicly managed competition is an auto race are not limited to the above.
 複数の目的関数が用いられる場合に、取得部11は、最適解の決定に用いる目的関数の選択結果を利用者端末装置20から取得してもよい。目的関数の選択結果は、例えば、レース結果の予測における最適解を利用する人物の操作によって利用者端末装置20に入力される目的関数の選択を、利用者端末装置20から取得してもよい。また、取得部11は、レース結果の予測における最適解を利用する人物の操作によって利用者端末装置20に入力される表示項目の選択を、利用者端末装置20から取得してもよい。 When a plurality of objective functions are used, the acquisition unit 11 may acquire from the user terminal device 20 the selection result of the objective function used to determine the optimal solution. The selection result of the objective function may be acquired from the user terminal device 20, for example, by inputting the selection of the objective function to the user terminal device 20 through the operation of a person who uses the optimal solution in predicting the race result. Further, the acquisition unit 11 may acquire from the user terminal device 20 a selection of display items that is input to the user terminal device 20 by an operation of a person who uses the optimal solution in predicting the race result.
 意思決定支援システム10がレース結果の予測における最適解の決定に用いる目的関数を生成する場合に、取得部11は、目的関数を生成するためのデータとして、公営競技のレースに関する意思決定履歴を取得してもよい。取得部11は、例えば、取得した公営競技のレースに関する意思決定履歴を記憶部15に保存する。 When the decision support system 10 generates an objective function to be used for determining the optimal solution in predicting race results, the acquisition unit 11 acquires the decision history regarding races in publicly managed competitions as data for generating the objective function. You may. The acquisition unit 11 stores, for example, the acquired decision-making history regarding races in public competitions in the storage unit 15.
 決定部12は、目的関数と、公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する。目的関数は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成される。決定部12は、例えば、目的関数と、公営競技のレースに関する情報を用いて最適化問題を解くことで、レース結果の予測における最適解を決定する。 The determining unit 12 determines the optimal solution for predicting race results using the objective function and information regarding races in publicly managed competitions. The objective function is generated in advance using inverse reinforcement learning based on the decision-making history related to predicting race results in publicly managed competitions. The determining unit 12 determines an optimal solution for predicting race results by solving an optimization problem using, for example, an objective function and information regarding races in publicly managed competitions.
 決定部12は、レース結果の予測における最適解を決定した理由を抽出してもよい。決定部12は、例えば、最適解の決定に対する影響が他の項目よりも大きい特徴量を最適解の決定の理由として抽出する。最適解の決定に対する影響が他の項目よりも大きい特徴量は、例えば、重み係数が他の特徴量よりも大きい特徴量である。また、決定部12は、最適解の決定に対する影響の度合いがあらかじめ設定された基準以上の特徴量を最適解の決定の理由として出力する。目的関数が複数の意図の集合として生成される場合に、決定部12は、最適解の決定に対する影響が他の意図より大きい意図に関する情報を最適解の決定の理由として出力してもよい。意図に関する情報は、例えば、意図に対応する数式または意図を示す単語である。また、目的関数が複数の意図の集合として生成される場合に、決定部12は、最適解の決定に対する影響があらかじめ設定された基準以上の意図に関する情報を最適解の決定の理由として出力してもよい。 The determining unit 12 may extract the reason for determining the optimal solution in predicting the race result. For example, the determining unit 12 extracts a feature quantity that has a larger influence on determining the optimal solution than other items as the reason for determining the optimal solution. A feature that has a greater influence on determining the optimal solution than other items is, for example, a feature that has a larger weighting coefficient than the other features. Further, the determining unit 12 outputs a feature amount whose degree of influence on determining the optimal solution is equal to or higher than a preset standard as a reason for determining the optimal solution. When the objective function is generated as a set of a plurality of intentions, the determining unit 12 may output information regarding intentions that have a greater influence on determining the optimal solution than other intentions as the reason for determining the optimal solution. The information regarding the intention is, for example, a mathematical formula corresponding to the intention or a word indicating the intention. Furthermore, when the objective function is generated as a set of multiple intentions, the determining unit 12 outputs information regarding intentions whose influence on determining the optimal solution is greater than a preset standard as the reason for determining the optimal solution. Good too.
 決定部12は、複数の目的関数のうちいずれかの目的関数を用いて、レース結果の予測における最適解を決定してもよい。決定部12は、例えば、レース結果の予測における最適解を利用する人物の属性に応じた目的関数を用いて、最適化結果を決定する。レース結果の予測における最適解を利用する人物の属性は、例えば、レース結果の予測における最適解を利用する人物の立場である。レース結果の予測における最適解を利用する人物の属性は、例えば、投票券の購入者、記者、解説者およびレースの編成担当の区分で設定される。レース結果の予測における最適解を利用する人物の属性は、初級者、中級者および上級者の区分で設定されてもよい。最適解を利用する人物の属性は、上記に限られない。また、決定部12は、意思決定履歴が用いられた人物ごとに生成された目的関数を用いて、レース結果の予測における最適解を決定してもよい。決定部12は、例えば、解説者Aの意思決定履歴によって生成された目的関数を用いるAモデルと、解説者Bの意思決定履歴によって生成された目的関数を用いるBモデルのいずれかを用いて、レース結果の予測における最適解を決定してもよい。 The determining unit 12 may determine the optimal solution for predicting the race result using any one of the plurality of objective functions. The determining unit 12 determines the optimization result using, for example, an objective function according to the attribute of the person who uses the optimal solution in predicting the race result. The attribute of the person who uses the optimal solution in predicting the race result is, for example, the position of the person who uses the optimal solution in predicting the race result. The attributes of the person who uses the optimal solution in predicting the race result are set, for example, by classification of voting ticket purchaser, reporter, commentator, and race organizer. The attributes of the person who uses the optimal solution in predicting the race result may be set for beginners, intermediates, and experts. The attributes of the person who uses the optimal solution are not limited to the above. Further, the determining unit 12 may determine the optimal solution for predicting the race result using an objective function generated for each person whose decision-making history is used. The determining unit 12 uses, for example, either model A that uses an objective function generated based on commentator A's decision-making history, or model B that uses an objective function generated based on commentator B's decision-making history, An optimal solution for predicting race results may be determined.
 レース結果の予測における最適解が購入する投票券の組み合わせである場合に、決定部12は、例えば、目的関数を用いて、払い戻し金額が高くなる投票券の組み合わせを決定する。決定部12は、例えば、購入額よりも払い戻し金額が高くなるように投票券の組み合わせを決定する。決定部12は、例えば、投票券を購入するための予算を制約条件として、購入額よりも払い戻し金額が高くなるように投票券の組み合わせを決定する。決定部12は、最適解として、複数のレースにおける投票券の組み合わせを決定してもよい。決定部12は、高いリターンを得るため、最適解として、オッズが基準以上の投票券から、投票券の組み合わせを決定してもよい。また、決定部12は、投機的な購入を防ぐため、最適解として、オッズが基準未満の投票券から、投票券の組み合わせを決定してもよい。 If the optimal solution in predicting the race result is a combination of voting tickets to purchase, the determining unit 12 uses, for example, an objective function to determine a combination of voting tickets that will result in a higher payout amount. For example, the determining unit 12 determines a combination of voting tickets such that the refund amount is higher than the purchase amount. For example, the determining unit 12 determines a combination of voting tickets such that the refund amount is higher than the purchase amount, using the budget for purchasing voting tickets as a constraint condition. The determining unit 12 may determine a combination of voting tickets for a plurality of races as the optimal solution. In order to obtain a high return, the determining unit 12 may determine a combination of voting tickets from among voting tickets with odds equal to or higher than a reference value as an optimal solution. Furthermore, in order to prevent speculative purchases, the determining unit 12 may determine a combination of voting tickets from among voting tickets whose odds are less than a standard as an optimal solution.
 決定部12は、最適解として、投票券を購入する対象のレースを決定してもよい。投票券を購入する対象のレースを決定する場合に、決定部12は、制約条件を満たすレースの中から投票券を購入する対象のレースを決定してもよい。決定部12は、例えば、最適解を利用する人物が収入を得てから一定期間内のレースから投票券を購入する対象のレースを決定する。決定部12は、最適解として、レースの着順を決定してもよい。最適解として決定するレース結果の予測は、上記の例に限られない。 The determining unit 12 may determine the race for which a voting ticket is to be purchased as the optimal solution. When determining a race for which a voting ticket is to be purchased, the determination unit 12 may determine a race for which a voting ticket is to be purchased from among races that satisfy the constraint conditions. The determining unit 12 determines, for example, a race for which a voting ticket is to be purchased from among races within a certain period of time after the person using the optimal solution earns income. The determining unit 12 may determine the order of race finish as the optimal solution. The prediction of the race result determined as the optimal solution is not limited to the above example.
 出力部13は、決定部12が決定した最適解を基に、レース結果の予測に関する情報を出力する。レース結果の予測に関する情報は、例えば、決定部12が決定した最適解、最適解の決定の理由、決定した最適解を用いて生成される情報の少なくとも1つである。レース結果の予測に関する情報は、上記に限られない。最適解が購入する投票券の組み合わせである場合に、決定した最適解を用いて生成される情報は、例えば、払い戻し額である。出力部13は、例えば、利用者端末装置20に、決定した最適解を出力する。出力部13は、決定した最適解とともに、最適解の決定の理由を出力してもよい。出力部13は、意思決定支援システム10と接続されている、図示しない表示装置に、決定した最適解を出力してもよい。また、出力部13は、レース結果の予測における最適解を配信するサーバに、決定した最適解を出力してもよい。 The output unit 13 outputs information regarding prediction of the race result based on the optimal solution determined by the determination unit 12. The information regarding the prediction of the race result is, for example, at least one of the optimal solution determined by the determining unit 12, the reason for determining the optimal solution, and information generated using the determined optimal solution. Information regarding prediction of race results is not limited to the above. When the optimal solution is a combination of voting tickets to be purchased, the information generated using the determined optimal solution is, for example, the payout amount. The output unit 13 outputs the determined optimal solution to the user terminal device 20, for example. The output unit 13 may output the reason for determining the optimal solution together with the determined optimal solution. The output unit 13 may output the determined optimal solution to a display device (not shown) connected to the decision support system 10. Further, the output unit 13 may output the determined optimal solution to a server that distributes the optimal solution in predicting the race result.
 出力部13は、目的関数または目的関数に含まれる特徴量の重み係数を出力してもよい。目的関数に含まれる特徴量の重み係数は、特徴量が対応する説明変数の係数である。また、出力部13は、目的関数に含まれる特徴量の重み係数を可視化して出力してもよい。出力部13は、例えば、目的関数に含まれる重み係数をグラフに表示することで可視化して出力する。 The output unit 13 may output the objective function or the weighting coefficient of the feature amount included in the objective function. The weighting coefficient of the feature included in the objective function is the coefficient of the explanatory variable to which the feature corresponds. Further, the output unit 13 may visualize and output the weighting coefficients of the feature amounts included in the objective function. The output unit 13 visualizes and outputs the weighting coefficients included in the objective function by displaying them on a graph, for example.
 レース結果の予測における最適解が購入する投票券の組み合わせである場合に、出力部13は、最適解として示す投票券の組み合わせとともに、当該組み合わせで投票券を購入した場合における収支の見込みを出力してもよい。 When the optimal solution in predicting the race result is a combination of voting tickets to be purchased, the output unit 13 outputs the combination of voting tickets indicated as the optimal solution as well as the expected income and expenditure when purchasing voting tickets with the combination. It's okay.
 公営競技が競馬で場合に、出力部13は、レース結果の予測における最適解とともに、最適解に含まれる出走馬の画像と、出走馬に関する情報とを出力してもよい。出力部13は、レース結果の予測における最適解とともに、最適解に含まれる出走馬の画像と、出走馬に関する情報にいずれか一方を出力してもよい。 When the publicly managed competition is a horse race, the output unit 13 may output the optimal solution for predicting the race result, as well as images of the runners included in the optimal solution and information regarding the runners. The output unit 13 may output either one of the optimal solution in predicting the race result, an image of the runner included in the optimal solution, and information regarding the runner.
 図3は、レース結果の予測における最適解の表示画面の例を示す。図3の表示画面の例では、公営競技が競馬の場合に、競馬の投票券を購入する際における購入する投票券の組み合わせが最適解として表示されている。図3の表示画面の例では、東京競馬場で行われる第9レースにおける購入する投票券の組み合わせの最適解が表示されている。図3の表示画面の例では、予算が1000円である場合の、購入する投票券の組み合わせの最適解が表示されている。図3の表示画面の例では、投票券の組み合わせの最適解として、投票券の種別と、馬番号または枠番号および購入額を用いて表示されている。図3の表示画面の例では、投票券ごとのオッズが表示されている。図3の表示画面の例では、投票券の組み合わせの最適解として、馬連3-5が600円、馬連3-9が200円、枠連2-3が200円の組み合わせが表示されている。 FIG. 3 shows an example of the display screen of the optimal solution in predicting the race result. In the example of the display screen in FIG. 3, when the publicly managed game is horse racing, the combination of voting tickets to be purchased when purchasing voting tickets for horse racing is displayed as the optimal solution. In the example of the display screen in FIG. 3, the optimal solution for the combination of voting tickets to be purchased for the ninth race to be held at the Tokyo Racecourse is displayed. In the example of the display screen in FIG. 3, the optimal combination of voting tickets to purchase is displayed when the budget is 1000 yen. In the example of the display screen in FIG. 3, the optimal solution for the combination of voting tickets is displayed using the type of voting ticket, horse number or slot number, and purchase amount. In the example of the display screen in FIG. 3, odds for each voting ticket are displayed. In the example of the display screen in FIG. 3, the combinations of 600 yen for horse group 3-5, 200 yen for horse group 3-9, and 200 yen for frame group 2-3 are displayed as the optimal combinations of voting tickets.
 図4は、図3の表示画面の例において、払い戻し額と、収支をさらに表示する表示画面の例を示す。図4の表示画面の例では、投票券の種別と、馬番号または枠番号、購入額およびオッズに加え、着順予想が当たった場合における払い戻し金額が払戻額として表示されている。図4の表示画面の例では、下段に出走馬の着順ごとの収支が表示されている。図4の表示画面の例の収支の表示は、1着が馬番号3の出走馬、2着が馬番号5の出走馬であった場合に、収支がプラス720円になることを示す。また、図4の表示画面の例の収支の表示は、1着が馬番号3の出走馬、2着が馬番号9の出走馬であった場合に、収支がプラス320円になることを示す。 FIG. 4 shows an example of a display screen that further displays the refund amount and income and expenditure in the example of the display screen of FIG. 3. In the example of the display screen in FIG. 4, in addition to the type of voting ticket, horse number or slot number, purchase amount, and odds, the refund amount in the event that the predicted finish is correct is displayed as the refund amount. In the example of the display screen shown in FIG. 4, the bottom row displays the income and expenditure for each horse placed in the race. The balance display in the example of the display screen in FIG. 4 shows that if the first place is horse number 3, and the second place is horse number 5, the balance will be an additional 720 yen. In addition, the balance display in the example of the display screen in Figure 4 shows that if 1st place is run by horse number 3 and 2nd place is run by horse number 9, the balance will be an additional 320 yen. .
 図5は、図3の表示画面の例において、最適解の決定の理由がさらに表示される表示画面の例を示す。図5の表示画面の例は、最適解の決定の理由が、理由の欄に表示されている。決定の理由として、例えば、最適解の決定への影響が他の特徴量よりも大きい特徴量が表示される。最適解の決定への影響が他の特徴量よりも大きい特徴量は、例えば、重み係数が他の特徴量よりも大きい特徴量である。図5の表示画面の例では、特徴量のうち、血統と、レースの距離が決定の理由として表示されている。図5の表示画面の例は、例えば、馬連3-5が最適解である理由が、出走馬の血統であることを示す。 FIG. 5 shows an example of a display screen in which the reason for determining the optimal solution is further displayed in the example of the display screen of FIG. 3. In the example of the display screen in FIG. 5, the reason for determining the optimal solution is displayed in the reason column. As the reason for the decision, for example, a feature amount that has a greater influence on determining the optimal solution than other feature amounts is displayed. A feature quantity that has a larger influence on determining the optimal solution than other feature quantities is, for example, a feature quantity whose weighting coefficient is larger than that of the other feature quantities. In the example of the display screen in FIG. 5, among the feature amounts, pedigree and race distance are displayed as reasons for the decision. The example of the display screen in FIG. 5 shows, for example, that the reason why horse race 3-5 is the optimal solution is the pedigree of the horses entering the race.
 図6は、図5の表示画面の例に加えて、最適解に含まれる出走馬の情報を表示する表示画面の例を示す。図5の表示画面の例では、レーン3の出走馬の画像と、出走馬に関連する情報が参考情報として表示されている。図5の表示画面の例では、1頭の出走馬の情報のみが表示されているが、出力部13は、複数の出走馬の情報を、最適化結果に付加して出力してもよい。また、出力部13は、例えば、表示画面上においてクリックまたはタップされた出走馬に対応する情報を出力してもよい。 In addition to the example of the display screen in FIG. 5, FIG. 6 shows an example of a display screen that displays information on the horses included in the optimal solution. In the example of the display screen in FIG. 5, an image of a horse running in lane 3 and information related to the horse running are displayed as reference information. In the example of the display screen in FIG. 5, only information on one runner is displayed, but the output unit 13 may add information on a plurality of runners to the optimization result and output the result. Further, the output unit 13 may output information corresponding to a runner clicked or tapped on the display screen, for example.
 図7は、図5の表示画面の例に加えて、最適解の決定に用いた目的関数を表示する表示画面の例を示す。図6の表示画面の例では、下段の予測式の欄に、最適解の決定に用いられた目的関数の式または目的関数に含まれる特徴量の重み係数が表示される。 In addition to the example of the display screen in FIG. 5, FIG. 7 shows an example of a display screen that displays the objective function used to determine the optimal solution. In the example of the display screen shown in FIG. 6, the expression of the objective function used to determine the optimal solution or the weighting coefficient of the feature quantity included in the objective function is displayed in the prediction expression column at the bottom.
 意思決定支援システム10において目的関数が生成される場合に、生成部14は、公営競技のレース結果の予測に関する意思決定履歴に基づいた逆強化学習によって生成する。目的関数は、公営競技のレースに関する情報からレース結果の予測における最適解を決定する。生成部14は、例えば、公営競技のレース結果の予測に関する意思決定履歴を逆強化学習によって学習することで目的関数を生成する。生成部14は、目的関数に説明変数として含まれる特徴量の組み合わせと、各特徴量の重み係数と、制約条件とを含む意思決定モデルとして目的関数を生成してもよい。 When the objective function is generated in the decision-making support system 10, the generation unit 14 generates it by inverse reinforcement learning based on the decision-making history regarding prediction of race results in public competitions. The objective function determines the optimal solution for predicting race results from information regarding races in public competitions. The generation unit 14 generates the objective function by learning, for example, a history of decision making regarding prediction of race results in publicly managed competitions by inverse reinforcement learning. The generation unit 14 may generate the objective function as a decision-making model including a combination of feature quantities included as explanatory variables in the objective function, weighting coefficients for each feature quantity, and constraint conditions.
 生成部14は、公営競技のレース結果の予測に関する意思決定履歴に含まれる行動データを最適解であるとして学習し、最適化指標すなわち目的関数を生成する。生成部14は、複数の線形式を生成し、線形式を選択するルールを学習することで目的関数を生成する。生成部14は、意思決定履歴を利用して、報酬を用いた最適化と、目的関数の重み係数の更新とを繰り返すことによって、意思決定履歴に対応する意図を反映する目的関数を生成する。 The generation unit 14 learns the behavioral data included in the decision-making history regarding prediction of race results in publicly managed competitions as the optimal solution, and generates an optimization index, that is, an objective function. The generation unit 14 generates an objective function by generating a plurality of linear forms and learning rules for selecting the linear forms. The generation unit 14 generates an objective function that reflects the intention corresponding to the decision history by repeatedly performing optimization using a reward and updating the weighting coefficient of the objective function using the decision history.
 目的関数を生成する際に、生成部14は、目的関数に含まれる特徴量の重み係数を仮の値を用いて設定する。生成部14は、目的関数を、意思決定履歴を用いて逆強化学習を行う際に用いる。重み係数の仮の値は、例えば、ランダムに設定される。そして、生成部14は、設定した重み係数を用いて目的関数を生成する。目的関数を生成すると、生成部14は、意思決定履歴の状態データと、目的関数とを用いて最適化問題を解き、最適解を決定する。最適解を決定すると、生成部14は、決定した最適解と、意思決定履歴の行動データとを比較する。生成部14は、比較結果を基に、所定のアルゴリズムを用いて決定した最適解と、意思決定履歴の行動データの差が小さくなるように重み係数を更新し、目的関数を生成する。そして、生成部14は、意思決定履歴の状態データと、生成した目的関数と決定した最適解とを用いて最適化問題を解き、最適解を決定する処理を繰り返す。生成部14は、決定した最適解と、意思決定履歴の行動データとの差が終了条件としてあらかじめ設定された基準を満たす場合に、学習を終了する。そして、生成部14は、例えば、記憶部15に、生成した目的関数を保存する。 When generating the objective function, the generation unit 14 sets the weighting coefficients of the feature amounts included in the objective function using temporary values. The generation unit 14 uses the objective function when performing inverse reinforcement learning using the decision history. The tentative values of the weighting coefficients are set randomly, for example. Then, the generation unit 14 generates an objective function using the set weighting coefficients. After generating the objective function, the generation unit 14 solves the optimization problem using the state data of the decision history and the objective function, and determines an optimal solution. After determining the optimal solution, the generation unit 14 compares the determined optimal solution with behavioral data of the decision-making history. The generation unit 14 updates the weighting coefficients based on the comparison results so that the difference between the optimal solution determined using a predetermined algorithm and the behavioral data of the decision history becomes small, and generates an objective function. The generation unit 14 then solves the optimization problem using the state data of the decision history, the generated objective function, and the determined optimal solution, and repeats the process of determining the optimal solution. The generation unit 14 ends the learning when the difference between the determined optimal solution and the behavioral data of the decision history satisfies a criterion set in advance as an end condition. Then, the generation unit 14 stores the generated objective function in the storage unit 15, for example.
 生成部14は、例えば、最大エントロピー逆強化学習によって、目的関数を生成する。生成部14は、スパース逆強化学習によって、目的関数を生成してもよい。スパース逆強化学習によって目的関数を生成する場合に、生成部14は、例えば、特徴量について、あらかじめ全ての候補に対して重みを付与し、学習結果に応じて特徴量を選び出すことで目的関数を生成する。また、生成部14は、複数の目的関数を属性に応じて選択可能なように配分してもよい。目的関数の生成方法は、例えば、国際公開第2021/130915号に記載されている。 The generation unit 14 generates the objective function by, for example, maximum entropy inverse reinforcement learning. The generation unit 14 may generate the objective function by sparse inverse reinforcement learning. When generating an objective function by sparse inverse reinforcement learning, the generation unit 14 generates the objective function by, for example, assigning weights to all candidates for feature quantities in advance and selecting feature quantities according to the learning results. generate. Further, the generation unit 14 may allocate a plurality of objective functions so that they can be selected according to attributes. A method for generating an objective function is described in, for example, International Publication No. 2021/130915.
 記憶部15は、例えば、決定部12が最適解の決定に用いる目的関数を保存する。複数の目的関数が用いられる場合には、記憶部15は、複数の目的関数を保存する。また、目的関数に含まれる特徴量の重み係数が更新された場合に、記憶部15は、更新された特徴量の重み係数を保存してもよい。また、意思決定支援システム10が目的関数を生成する場合に、記憶部15は、目的関数の生成に用いる、過去のレースの予測に対しての意思決定履歴を保存してもよい。上記の各データは、記憶部15以外の記憶手段に保存されていてもよい。 The storage unit 15 stores, for example, an objective function that the determination unit 12 uses to determine the optimal solution. When a plurality of objective functions are used, the storage unit 15 stores the plurality of objective functions. Further, when the weighting coefficient of the feature quantity included in the objective function is updated, the storage unit 15 may store the updated weighting coefficient of the feature quantity. Furthermore, when the decision support system 10 generates an objective function, the storage unit 15 may store decision history for past race predictions, which is used to generate the objective function. Each of the above data may be stored in a storage means other than the storage unit 15.
 利用者端末装置20は、意思決定支援システム10から、レース結果の予測における最適解を取得する。そして、利用者端末装置20は、例えば、図示しない表示装置に、レース結果の予測における最適解を出力する。 The user terminal device 20 acquires the optimal solution for predicting the race result from the decision support system 10. Then, the user terminal device 20 outputs the optimal solution for predicting the race result to, for example, a display device (not shown).
 利用者によって、最適解を決定する目的関数の選択が行われる場合に、利用者端末装置20は、例えば、利用者の操作によって入力される目的関数の選択の入力を取得する。利用者端末装置20は、利用者の操作によって入力される目的関数の名称を取得してもよい。そして、利用者端末装置20は、意思決定支援システム10に、入力された目的関数の選択に関する入力結果を出力する。 When the user selects an objective function to determine the optimal solution, the user terminal device 20 obtains an input for selecting the objective function input by the user's operation, for example. The user terminal device 20 may acquire the name of the objective function input by the user's operation. The user terminal device 20 then outputs the input result regarding the selection of the input objective function to the decision support system 10.
 利用者端末装置20には、例えば、スマートフォン、タブレット型コンピュータ、ノート型コンピュータまたはデスクトップ型コンピュータが用いられる。利用者端末装置20に用いられる端末装置は、上記の例に限られない。 For example, a smartphone, a tablet computer, a notebook computer, or a desktop computer is used as the user terminal device 20. The terminal device used for the user terminal device 20 is not limited to the above example.
 情報管理サーバ30は、例えば、公営競技のレースに関する情報を保存または管理するサーバである。情報管理サーバ30は、公営競技のレースに関する情報の内容に応じて設置されている複数のサーバであってもよい。公営競技のレースに関する情報は、情報管理サーバ30が管理する記憶装置に保存されてもよい。 The information management server 30 is, for example, a server that stores or manages information regarding races in publicly managed competitions. The information management server 30 may be a plurality of servers installed according to the content of information regarding races in public competitions. Information regarding races in public competitions may be stored in a storage device managed by the information management server 30.
 意思決定支援システム10において、公営競技のレース結果の予測における最適解を決定する際の動作について説明する。図8は、意思決定支援システム10がレース結果の予測における最適解を決定する際の動作フローの例を示す図である。 The operation of the decision support system 10 when determining the optimal solution for predicting race results of publicly managed competitions will be explained. FIG. 8 is a diagram showing an example of an operation flow when the decision support system 10 determines the optimal solution in predicting the race result.
 取得部11は、公営競技のレースに関する情報を取得する(ステップS11)。取得部11は、例えば、情報管理サーバ30から、公営競技のレースに関する情報を取得する。 The acquisition unit 11 acquires information regarding races in publicly managed competitions (step S11). The acquisition unit 11 acquires information regarding races in publicly managed competitions, for example, from the information management server 30.
 公営競技のレースに関する情報が取得されると、決定部12は、予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する(ステップS12)。目的関数は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で生成される。決定部12は、例えば、目的関数と、取得した公営競技のレースに関する情報とを用いて最適化問題を解くことで、レース結果の予測における最適解を決定する。 When the information regarding the race in the publicly managed competition is acquired, the determining unit 12 uses the objective function generated in advance and the acquired information regarding the race in the publicly managed competition to determine the optimal solution for predicting the race result (step S12). The objective function is generated using inverse reinforcement learning based on the decision-making history related to predicting race results in public competitions. The determining unit 12 determines the optimal solution for predicting the race result by solving an optimization problem using, for example, the objective function and the acquired information regarding the publicly managed race.
 レース結果の予測における最適解が決定されると、出力部13は、決定部12が決定した最適解を出力する(ステップS13)。出力部13は、例えば、利用者端末装置20に、レース結果の予測における最適解を出力する。 Once the optimal solution for predicting the race result is determined, the output unit 13 outputs the optimal solution determined by the determining unit 12 (step S13). The output unit 13 outputs the optimal solution for predicting the race result to the user terminal device 20, for example.
 レース結果の予測における最適解を取得すると、利用者端末装置20は、例えば、図示しない表示装置に、レース結果の予測における最適解を出力する。 After acquiring the optimal solution for predicting the race result, the user terminal device 20 outputs the optimal solution for predicting the race result, for example, to a display device (not shown).
 意思決定支援システム10において、公営競技のレース結果の予測における最適解を決定する目的関数を生成する場合に、目的関数を生成する動作について説明する。図9は、意思決定支援システム10が目的関数を生成する際の動作フローの例を示す図である。意思決定支援システム10は、生成した目的関数を、公営競技のレース結果の予測における最適解を決定する際に用いる。 In the decision-making support system 10, an operation for generating an objective function will be described when generating an objective function to determine the optimal solution in predicting the race result of a publicly managed competition. FIG. 9 is a diagram illustrating an example of an operation flow when the decision support system 10 generates an objective function. The decision support system 10 uses the generated objective function when determining the optimal solution for predicting race results in public competitions.
 取得部11は、過去に実施されたレースにおける、公営競技のレース結果の予測に関する意思決定履歴を取得する(ステップS21)。意思決定履歴が取得されると、生成部14は、目的関数に含まれる特徴量の重み係数を仮の値を用いて設定する。生成部14は、目的関数を、意思決定履歴を用いて逆強化学習を行う際に用いる。仮の値を用いて重み係数を設定すると、生成部14は、設定した重み係数を用いて目的関数を生成する(ステップS22)。 The acquisition unit 11 acquires a history of decision-making regarding prediction of race results in publicly managed competitions in races held in the past (step S21). When the decision history is acquired, the generation unit 14 sets weighting coefficients of the feature amounts included in the objective function using temporary values. The generation unit 14 uses the objective function when performing inverse reinforcement learning using the decision history. After setting the weighting coefficient using the temporary value, the generation unit 14 generates an objective function using the set weighting coefficient (step S22).
 目的関数を生成すると、生成部14は、意思決定履歴と、目的関数とを用いて最適化問題を解き、最適解を決定する(ステップS23)。最適解を決定すると、生成部14は、決定した最適解と、意思決定履歴の行動データとを比較する(ステップS24)。 After generating the objective function, the generation unit 14 solves the optimization problem using the decision history and the objective function, and determines the optimal solution (step S23). After determining the optimal solution, the generation unit 14 compares the determined optimal solution with behavioral data of the decision-making history (step S24).
 決定した最適解があらかじめ設定された終了条件を満たす場合(ステップS25でYes)、生成部14は、生成した目的関数を保存する(ステップS26)。生成部14は、例えば、記憶部15に、生成した目的関数を保存する。保存された目的関数は、決定部12において最適解を決定する際に用いられる。 If the determined optimal solution satisfies the preset termination condition (Yes in step S25), the generation unit 14 stores the generated objective function (step S26). The generation unit 14 stores the generated objective function in the storage unit 15, for example. The stored objective function is used by the determining unit 12 to determine the optimal solution.
 決定した最適解が終了条件を満たさない場合(ステップS25でNo)、生成部14は、あらかじめ設定されたアルゴリズムを用いて特徴量の重み係数を更新する(ステップS27)。特徴量の重み係数を更新すると、生成部14は、更新した重み係数を用いて、目的関数を生成する(ステップS28)。目的関数を生成すると、生成部14は、ステップS23からの処理を繰り返して行う。 If the determined optimal solution does not satisfy the termination condition (No in step S25), the generation unit 14 updates the weighting coefficient of the feature amount using a preset algorithm (step S27). After updating the weighting coefficients of the feature amounts, the generation unit 14 generates an objective function using the updated weighting coefficients (step S28). After generating the objective function, the generation unit 14 repeats the processing from step S23.
 本実施形態の意思決定支援システム10は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する。よって、意思決定支援システム10を用いることで、意思決定履歴に基づくレース結果の予測における最適解を容易に得られるようになる。このため、意思決定支援システム10を用いることで、公営競技のレース結果に関する意思決定を容易に行うことができる。 The decision-making support system 10 of this embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in publicly managed competitions, and information regarding races in publicly managed competitions. Determine the optimal solution in prediction. Therefore, by using the decision-making support system 10, it becomes possible to easily obtain an optimal solution for predicting the race result based on the decision-making history. Therefore, by using the decision support system 10, it is possible to easily make decisions regarding the race results of publicly managed competitions.
 また、投票券の購入による収支の情報を出力する場合に、意思決定支援システム10は、例えば、最適解の決定を基にした投票券の組み合わせごとの収支の情報を出力することで、最適化の決定の結果を適用するかについての利用者による判断を、より容易することができる。 In addition, when outputting information on the income and expenditure resulting from the purchase of voting tickets, the decision support system 10 can perform optimization by, for example, outputting information on the income and expenditure for each combination of voting tickets based on the determination of the optimal solution. This makes it easier for the user to decide whether to apply the result of the decision.
 また、複数の目的関数を用いる場合に、意思決定支援システム10は、例えば、レース結果の予測における最適解を利用する人物に応じた目的関数を用いることで、レース結果の予測における最適解を利用する人物に適する最適解を出力することができる。 In addition, when using a plurality of objective functions, the decision support system 10 uses the optimal solution in predicting the race result by using an objective function depending on the person who uses the optimal solution in predicting the race result, for example. It is possible to output the optimal solution suitable for the person who uses the software.
 (第2の実施形態)
 本発明の第2の実施形態について図を参照して詳細に説明する。図10は、本実施形態の情報処理システムの例を示す図である。一例として、情報処理システムは、意思決定支援システム40と、利用者端末装置20と、情報管理サーバ30を備える。意思決定支援システム40は、ネットワークを介して、利用者端末装置20と接続する。また、意思決定支援システム40は、ネットワークを介して、情報管理サーバ30と接続する。利用者端末装置20と、情報管理サーバ30の数は、適宜、設定される。本実施形態の利用者端末装置20と、情報管理サーバ30の機能は、第1の実施形態の利用者端末装置20と、情報管理サーバ30と同様である。
(Second embodiment)
A second embodiment of the present invention will be described in detail with reference to the drawings. FIG. 10 is a diagram showing an example of the information processing system of this embodiment. As an example, the information processing system includes a decision support system 40, a user terminal device 20, and an information management server 30. The decision support system 40 is connected to the user terminal device 20 via a network. Further, the decision support system 40 is connected to the information management server 30 via a network. The numbers of user terminal devices 20 and information management servers 30 are set as appropriate. The functions of the user terminal device 20 and the information management server 30 of this embodiment are the same as those of the user terminal device 20 and the information management server 30 of the first embodiment.
 第1の実施形態の意思決定支援システム10は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する。そして、意思決定支援システム10は、公営競技のレース結果の予測における最適解を出力する。このような構成に加え、本実施形態の意思決定支援システム40は、例えば、レース結果の予測における最適解を決定するために用いる目的関数に含まれる特徴量の重み係数の変更が可能である。 The decision-making support system 10 of the first embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in publicly managed competitions, and acquired information regarding races in publicly managed competitions. , determine the optimal solution in predicting race results. The decision support system 10 then outputs the optimal solution for predicting the race result of the publicly managed competition. In addition to such a configuration, the decision support system 40 of the present embodiment can, for example, change the weighting coefficient of the feature amount included in the objective function used to determine the optimal solution in predicting the race result.
 意思決定支援システム40は、例えば、目的関数に含まれる特徴量の重み係数の組み合わせに対応する、単語の埋め込み表現を可視化して出力する。単語は、例えば、公営競技のレース結果の予測に関する意思決定における傾向を示す。単語は、例えば、意図に対応する。また、意思決定支援システム40は、例えば、目的関数に含まれる特徴量の重み係数の組み合わせに対応する、単語の埋め込み表現を含む文章を検索する。意思決定支援システム40は、例えば、単語の埋め込み表現に対応する目的関数に含まれる特徴量の重み係数の組み合わせを用いて目的関数を更新する。 For example, the decision support system 40 visualizes and outputs embedded expressions of words that correspond to combinations of weighting coefficients of feature quantities included in the objective function. The words indicate, for example, trends in decision-making regarding prediction of race results in public competitions. A word corresponds to an intention, for example. Further, the decision support system 40 searches for a sentence that includes an embedded expression of a word that corresponds to a combination of weighting coefficients of feature quantities included in the objective function, for example. The decision support system 40 updates the objective function using, for example, a combination of weighting coefficients of features included in the objective function corresponding to the embedded representation of the word.
 意思決定支援システム40の構成について説明する。図11は、意思決定支援システム40の構成の例を示す図である。意思決定支援システム40は、取得部11と、決定部12と、出力部41と、生成部14と、記憶部15と、受付部42と、更新部43と、係数予測部44と、表現予測部45と、クエリ生成部46と、検索部47を備える。 The configuration of the decision support system 40 will be explained. FIG. 11 is a diagram showing an example of the configuration of the decision support system 40. As shown in FIG. The decision support system 40 includes an acquisition unit 11, a determination unit 12, an output unit 41, a generation unit 14, a storage unit 15, a reception unit 42, an update unit 43, a coefficient prediction unit 44, and an expression prediction unit. 45, a query generation section 46, and a search section 47.
 意思決定支援システム40の取得部11、決定部12、生成部14および記憶部15の構成と機能は、第1の実施形態の意思決定支援システム10の取得部11、決定部12、出力部13、生成部14および記憶部15とそれぞれ同様である。 The configuration and functions of the acquisition unit 11, determination unit 12, generation unit 14, and storage unit 15 of the decision support system 40 are the same as the acquisition unit 11, determination unit 12, and output unit 13 of the decision support system 10 of the first embodiment. , the generation unit 14 and the storage unit 15, respectively.
 出力部41は、例えば、第1の実施形態の出力部13が行う処理に加え、例えば、目的関数の重み係数の変更に関する画面を出力する。また、出力部41は、例えば、検索クエリを用いて検索を行った検索結果を出力する。 In addition to the processing performed by the output unit 13 of the first embodiment, the output unit 41 outputs, for example, a screen related to changing the weighting coefficient of the objective function. Further, the output unit 41 outputs, for example, a search result obtained by performing a search using a search query.
 出力部41は、例えば、目的関数の重み係数の変更に関する画面として、目的関数の重み係数の入力欄を有する表示画面を出力する。出力部41は、目的関数の重み係数をグラフとして表示し、グラフ上で重み係数を変更の操作を行うための表示画面を出力してもよい。 The output unit 41 outputs, for example, a display screen having an input field for the weighting coefficient of the objective function as a screen related to changing the weighting coefficient of the objective function. The output unit 41 may display the weighting coefficients of the objective function as a graph and output a display screen for changing the weighting coefficients on the graph.
 目的関数の重み係数をグラフとして出力する場合に、出力部41は、例えば、特徴量のうち、目的関数による最適解の決定に与える寄与が大きい特徴量を軸としたグラフを出力する。出力部41は、例えば、特徴量のうち、目的関数による最適解の決定に与える寄与が大きい2つの特徴量を軸とした2次元のグラフを出力する。また、出力部41は、例えば、特徴量のうち、目的関数による最適解の決定への影響が対照的な特徴量を軸としたグラフを出力してもよい。目的関数による最適解の決定への影響が対照的な特徴量とは、例えば、着順の予測を行う場合に、着順が高くなる予測への影響が大きい特徴量と、着順が低くなる予測への影響が大きい特徴量の組み合わせをいう。また、出力部41は、特徴量のうち、目的関数による最適解の決定に与える寄与が大きい3つ以上の特徴量を軸とした3次元以上のグラフを出力してもよい。また、出力部41は、多次元の特徴量を次元削減処理によって2次元化した結果をグラフとして表示してもよい。 When outputting the weighting coefficients of the objective function as a graph, the output unit 41 outputs, for example, a graph centered on feature quantities that make a large contribution to determining the optimal solution by the objective function among the feature quantities. The output unit 41 outputs, for example, a two-dimensional graph centered on two feature quantities that make a large contribution to determining the optimal solution by the objective function among the feature quantities. Further, the output unit 41 may output, for example, a graph centered on feature quantities that have contrasting effects on the determination of the optimal solution by the objective function among the feature quantities. Features that have contrasting effects on determining the optimal solution by the objective function are, for example, when predicting the order of finish, features that have a large effect on prediction that will result in a high order of finish, and features that have a large effect on the prediction that will result in a low order of finish. A combination of features that has a large impact on prediction. Furthermore, the output unit 41 may output a three-dimensional or more-dimensional graph centered on three or more feature quantities that greatly contribute to determining the optimal solution by the objective function among the feature quantities. Further, the output unit 41 may display the result of converting the multidimensional feature amount into two dimensions through dimension reduction processing as a graph.
 出力部41は、例えば、検索クエリを用いて文章を検索した結果を出力する。検索の対象となる文章は、例えば、公営競技のレース結果の予測に関係する文章である。公営競技が競売である場合に、検索の対象となる文章は、例えば、「東京9レースでやまっけだして勝負したら、大勝ちした。」のような文章である。」出力部41は、例えば、利用者端末装置20に、検索部47が検索した検索クエリに対応する表現を含む文章を出力する。検索部47がSNSの投稿を検索対象とする場合に、出力部41は、検索部47が検索した検索クエリに対応する表現を含む投稿を出力する。 The output unit 41 outputs, for example, the result of searching for a text using a search query. The text to be searched is, for example, a text related to predicting race results in publicly managed competitions. When the publicly managed competition is an auction, the sentence to be searched is, for example, a sentence such as ``I started racing in Tokyo 9 races and won big.'' ” The output unit 41 outputs, to the user terminal device 20, for example, a sentence including an expression corresponding to the search query searched by the search unit 47. When the search unit 47 searches for posts on SNS, the output unit 41 outputs posts that include expressions corresponding to the search query searched by the search unit 47.
 受付部42は、例えば、目的関数に含まれる特徴量の重み係数を変更する変更値を受け付ける。受付部42は、例えば、利用者端末装置20から、目的関数に含まれる特徴量の重み係数を変更する変更値を受け付ける。目的関数に含まれる特徴量の重み係数を変更する変更値は、例えば、利用者端末装置20の利用者によって、利用者端末装置20に入力される。受付部42は、意思決定支援システム40に接続された、図示しない入力装置から変更値を受け付けてもよい。 The receiving unit 42 receives, for example, a change value for changing the weighting coefficient of the feature quantity included in the objective function. The receiving unit 42 receives, for example, a change value for changing the weighting coefficient of the feature included in the objective function from the user terminal device 20. A change value for changing the weighting coefficient of the feature quantity included in the objective function is input into the user terminal device 20 by the user of the user terminal device 20, for example. The reception unit 42 may receive the change value from an input device (not shown) connected to the decision support system 40.
 目的関数に含まれる特徴量の重み係数がグラフとして出力される場合に、受付部42は、グラフ上に表示される特徴量の重み係数を変更することで入力される変更値を受け付けてもよい。受付部42は、例えば、特徴量の重み係数のグラフ上の表示位置を変更することで入力される変更値を受け付けてもよい。また、受付部42は、特徴量の重み係数を示すリストにおいて、表示される順番を変更することで入力される変更値を受け付けてもよい。 When the weighting coefficients of the feature quantities included in the objective function are output as a graph, the reception unit 42 may accept changed values input by changing the weighting coefficients of the feature quantities displayed on the graph. . For example, the receiving unit 42 may receive a change value input by changing the display position of the weighting coefficient of the feature amount on the graph. Further, the reception unit 42 may receive a change value input by changing the display order in a list indicating weighting coefficients of feature amounts.
 単語の埋め込み表現に対応する目的関数に含まれる特徴量の重み係数の組み合わせを用いて目的関数を更新する場合に、受付部42は、利用者が選択する、特徴量の重み係数の組み合わせを受け付け。受付部42は、利用者が選択する、特徴量の重み係数の組み合わせに対応する意図を受け付けてもよい。受付部42は、例えば、利用者の操作によって利用者端末装置20に入力される、意図の選択結果を、利用者端末装置20から取得することで意図を受け付ける。 When updating the objective function using a combination of weighting coefficients of feature quantities included in the objective function corresponding to the embedded representation of a word, the receiving unit 42 accepts the combination of weighting coefficients of feature quantities selected by the user. . The reception unit 42 may receive an intention corresponding to a combination of weighting coefficients of feature amounts selected by the user. The reception unit 42 receives the intention by acquiring, from the user terminal device 20, the selection result of the intention, which is input into the user terminal device 20 by the user's operation, for example.
 更新部43は、例えば、目的関数に含まれる特徴量の重み係数の変更値が受け付けられた場合に、取得された値を用いて、目的関数に含まれる特徴量の重み係数を更新する。更新部43は、例えば、重み係数を更新した目的関数を記憶部15に保存する。 For example, when a change value of the weighting coefficient of the feature quantity included in the objective function is received, the updating unit 43 updates the weighting coefficient of the feature quantity included in the objective function using the obtained value. The updating unit 43 stores, for example, the objective function with updated weighting coefficients in the storage unit 15.
 受け付けられた意図に基づいて目的関数を更新する場合に、更新部43は、例えば、受付部42が受け付けた意図それぞれに対応する目的関数を抽出する。そして、更新部43は、抽出した目的関数に含まれる特徴量の重み係数から、特徴量ごとの重み係数の平均値を算出する。更新部43は、算出した重み係数それぞれの平均値を、特徴量それぞれの重み係数として用いて目的関数を更新する。更新部43は、複数の目的関数に含まれる特徴量の重み係数から、新たな重み係数を算出する際に、平均値以外の値を用いて重み係数を算出してもよい。更新部43は、例えば、重視する意図に対応する目的関数に含まれる特徴量の重み係数の影響が大きくなるように新たな重み係数を算出してもよい。更新された目的関数は、決定部12によるレース結果の予測の最適解の決定に用いられる。 When updating the objective function based on the accepted intentions, the updating unit 43 extracts, for example, an objective function corresponding to each intention accepted by the accepting unit 42. Then, the updating unit 43 calculates the average value of the weighting coefficients for each feature quantity from the weighting coefficients of the feature quantities included in the extracted objective function. The updating unit 43 updates the objective function using the average value of each of the calculated weighting coefficients as a weighting coefficient of each feature amount. When calculating a new weighting coefficient from the weighting coefficients of the feature amounts included in the plurality of objective functions, the updating unit 43 may calculate the weighting coefficient using a value other than the average value. For example, the updating unit 43 may calculate a new weighting coefficient so that the influence of the weighting coefficient of the feature amount included in the objective function corresponding to the intention to be emphasized becomes greater. The updated objective function is used by the determining unit 12 to determine the optimal solution for predicting the race result.
 入力された意図に基づいて目的関数を更新する場合に、更新部43は意図を示す単語の埋め込み表現に対応する特徴量の重み係数の組み合わせを用いて目的関数を更新してもよい。更新部43は、例えば、係数予測部44が予測する、意図を示す単語の埋め込み表現に対応する特徴量の重み係数の組み合わせを用いて目的関数を更新する。 When updating the objective function based on the input intention, the updating unit 43 may update the objective function using a combination of weighting coefficients of features corresponding to the embedded expression of the word indicating the intention. The updating unit 43 updates the objective function using, for example, a combination of weighting coefficients of feature amounts corresponding to the embedded expression of the word indicating the intention, which is predicted by the coefficient predicting unit 44.
 図12は、図3のレース結果の予測における最適解の表示画面において、目的関数の式を表示し、重み係数の変更の入力を受け付ける表示画面の例を示す。図12の表示画面の例では、予測式の欄に目的関数の式が表示される。図12の表示画面の例では、目的関数の重み係数の変更は、例えば、利用者の操作によって目的関数の重み係数の式が書き換えられ、更新ボタンが押されることで行われる。 FIG. 12 shows an example of a display screen that displays the objective function equation and accepts input for changing the weighting coefficient on the display screen of the optimal solution in predicting the race result in FIG. 3. In the example of the display screen in FIG. 12, the objective function equation is displayed in the prediction equation column. In the example of the display screen in FIG. 12, the weighting coefficient of the objective function is changed by, for example, rewriting the expression of the weighting coefficient of the objective function by a user's operation and pressing an update button.
 図13は、目的関数に含まれる特徴量の重み係数を2次元のグラフ上に表示する表示画面の例を示す。図13の表示画面の例は、目的関数に含まれる特徴量のうち、xとxの重み係数をグラフとして示す。図13の表示画面の例の「現在の意図」は、最適解の決定に用いている目的関数に含まれる特徴量の重み係数を示す。図13の表示画面の例の「Aさんモデルの意図」は、Aさんの意思決定履歴を用いて生成された目的関数に含まれる特徴量の重み係数を示す。また、図13の表示画面の例の「Bさんモデルの意図」は、Bさんの意思決定履歴を用いて生成された目的関数に含まれる特徴量の重み係数を示す。 FIG. 13 shows an example of a display screen that displays weighting coefficients of feature quantities included in the objective function on a two-dimensional graph. The example display screen in FIG. 13 shows the weighting coefficients of x n and x m among the feature amounts included in the objective function as a graph. "Current intention" in the example display screen of FIG. 13 indicates the weighting coefficient of the feature quantity included in the objective function used to determine the optimal solution. "Intention of Mr. A's model" in the example of the display screen in FIG. 13 indicates the weighting coefficient of the feature amount included in the objective function generated using Mr. A's decision-making history. Further, "intention of Mr. B model" in the example of the display screen in FIG. 13 indicates the weighting coefficient of the feature quantity included in the objective function generated using Mr. B's decision-making history.
 目的関数に含まれる特徴量の重み係数を変更する場合に、図13の表示画面の例において、例えば、「現在の意図」のグラフ上の位置を、マウス操作によってドラッグして変更することで、移動先の位置に対応する重み係数に変更されるようにしてもよい。 When changing the weighting coefficient of the feature quantity included in the objective function, in the example of the display screen in FIG. 13, for example, by changing the position of "Current Intention" on the graph by dragging it with the mouse, The weighting coefficient may be changed to correspond to the destination position.
 図14は、レース結果の予測における最適解の決定への寄与が大きい特徴量を、影響の大きさ順にリストとして表示する表示画面の例を示す。図14の表示画面の例において、例えば、マウス操作によって特徴量の並び順を変えて更新ボタンが押されることで、重み係数の値が変更される。更新部43は、例えば、2つの特徴量の位置が変更された場合に、2つの特徴量の重み係数の値を入れ替えることで特徴量の重み係数を変更する。 FIG. 14 shows an example of a display screen that displays feature quantities that make a large contribution to determining the optimal solution in predicting race results as a list in order of magnitude of influence. In the example of the display screen in FIG. 14, the value of the weighting coefficient is changed by, for example, changing the arrangement order of the feature amounts by operating a mouse and pressing an update button. For example, when the positions of the two feature quantities are changed, the updating unit 43 changes the weighting coefficients of the two feature quantities by exchanging the values of the weighting coefficients of the two feature quantities.
 図15は、2つの意図に対応する特徴量の重み係数をグラフとして示す表示画面の例である。図15の表示画面の例では、上端と後段でそれぞれ異なる目的関数に含まれる特徴量の重み係数を示す。図15の表示画面の例において、例えば、マウス操作によって特徴量のグラフの高さをドラッグすることで変更した後、更新ボタンが押されることで、特徴量の重み係数は、特徴量のグラフの高さに対応する値に変更される。更新部43は、例えば、変更後のグラフの高さ応じた値を用いて特徴量の重み係数を変更する。 FIG. 15 is an example of a display screen that shows weighting coefficients of feature amounts corresponding to two intentions as a graph. In the example of the display screen in FIG. 15, weighting coefficients of features included in different objective functions are shown at the upper end and at the latter stage. In the example of the display screen in Figure 15, for example, by changing the height of the feature graph by dragging the mouse, and then pressing the update button, the weighting coefficient of the feature can be changed by dragging the height of the feature graph. Changed to the value corresponding to the height. The updating unit 43 changes the weighting coefficient of the feature amount using, for example, a value that corresponds to the height of the changed graph.
 係数予測部44は、例えば、単語の埋め込み表現に対応する特徴量の重み係数の組み合わせを予測する。係数予測部44は、単語の埋め込み表現と特徴量の重み係数の組みわせとの関係を学習した係数予測モデルに基づいて、埋め込み表現に対応する重み係数の組み合わせを予測する。係数予測モデルは、例えば、自然言語処理モデルを用いてベクトルに変換した単語の埋め込み表現と、特徴量の重み係数の組みわせとの関係を学習した学習済みモデルである。自然言語処理モデルとしては、例えば、BERT(Bidirectional Encoder Representations from Transformers)を用いることができる。 The coefficient prediction unit 44 predicts, for example, a combination of weighting coefficients of feature amounts corresponding to the embedded representation of a word. The coefficient prediction unit 44 predicts a combination of weighting coefficients corresponding to an embedded expression based on a coefficient prediction model that has learned the relationship between an embedded expression of a word and a combination of weighting coefficients of a feature amount. The coefficient prediction model is, for example, a trained model that has learned the relationship between the embedded representation of a word converted into a vector using a natural language processing model and the combination of weighting coefficients of feature amounts. As the natural language processing model, for example, BERT (Bidirectional Encoder Representations from Transformers) can be used.
 表現予測部45は、例えば、目的関数に含まれる特徴量の重み係数に対応する表現を予測する。表現予測部45は、例えば、目的関数に含まれる特徴量の重み係数に対応する単語の埋め込み表現を予測する。表現予測部45は、重み係数の組み合わせと、単語の埋め込み表現とのを学習した表現予測モデルに基づいて、重み係数の組み合わせに対応する埋め込み表現を予測する。重み係数の組み合わせに対応する埋め込み表現は、重み係数の組み合わせに対応する埋め込み表現と類似する表現を含む。 For example, the expression prediction unit 45 predicts an expression corresponding to the weighting coefficient of the feature amount included in the objective function. The expression prediction unit 45 predicts, for example, an embedded expression of a word corresponding to a weighting coefficient of a feature included in the objective function. The expression prediction unit 45 predicts the embedded expression corresponding to the combination of weighting coefficients based on the expression prediction model that has learned the combination of weighting coefficients and the embedded expression of the word. The embedded representation corresponding to the combination of weighting coefficients includes representations similar to the embedded representation corresponding to the combination of weighting coefficients.
 図16は、意図空間と、言語空間の対応を模式的に示す図である。意図空間は、目的関数に含まれる特徴量によって表される空間である。言語空間は、単語の埋め込み表現によって表される空間である。言語空間は、言語の意味的な類似度が連続値として定義される空間である。 FIG. 16 is a diagram schematically showing the correspondence between intention space and language space. The intention space is a space represented by the feature amounts included in the objective function. Linguistic space is a space represented by embedded representations of words. A linguistic space is a space in which the semantic similarity of languages is defined as a continuous value.
 図16の例において、意図空間のグラフの縦軸と横軸は、例えば、目的関数に含まれる特徴量の重み係数から選択された特徴量を用いて設定される。図16の例において、言語空間のグラフの縦軸と横軸は、言語上の意味を基に設定される。言語空間のグラフで近傍にある表現どうしは、意味が類似した表現である。また、図16の例において、意図空間における意図Aは、言語空間における「勝負師」に対応するとする。また、図16の例において、意図空間における意図Bは、言語空間における「逆張り」に対応するとする。表現予測部45は、これらの組み合わせを訓練データとする機械学習によって生成された予測モデルを用いて、特徴量の重み係数に対応する単語の埋め込み表現を予測する。 In the example of FIG. 16, the vertical and horizontal axes of the graph of the intention space are set using, for example, feature quantities selected from the weighting coefficients of the feature quantities included in the objective function. In the example of FIG. 16, the vertical and horizontal axes of the language space graph are set based on linguistic meaning. Expressions that are close to each other in the graph of language space are expressions that have similar meanings. Furthermore, in the example of FIG. 16, it is assumed that intention A in the intention space corresponds to "gamer" in the language space. Furthermore, in the example of FIG. 16, it is assumed that intention B in the intention space corresponds to "contrarian" in the language space. The expression prediction unit 45 uses a prediction model generated by machine learning using these combinations as training data to predict the embedded expression of the word corresponding to the weighting coefficient of the feature amount.
 図17は、表現予測モデルを用いた予測時における意図空間と、言語空間の対応を模式的に示す図である。図17の例において、意図空間における意図Cは、言語空間における「やまっけ」に対応するとする。このとき、表現予測部45が用いる表現予測モデルは、意図Cの特徴量の組み合わせに対し、言語空間における「やまっけ」を予測結果とする。 FIG. 17 is a diagram schematically showing the correspondence between the intention space and the language space during prediction using the expression prediction model. In the example of FIG. 17, it is assumed that intention C in the intention space corresponds to "yamakke" in the language space. At this time, the expression prediction model used by the expression prediction unit 45 predicts "Yamake" in the language space for the combination of feature amounts of intention C.
 図18は、表現予測モデルを用いた予測時における意図空間と、言語空間の対応を、言語空間における類似範囲を含めて模式的に示す図である。図18の例において、意図空間における意図Cは、言語空間における「やまっけ」に対応するとする。また、「勝負師」が「やまっけ」の類似範囲に含まれるとする。このとき、表現予測モデルは、意図Cの特徴量の組み合わせに対し、言語空間における「やまっけ」と、「勝負師」とを予測結果としてもよい。 FIG. 18 is a diagram schematically showing the correspondence between the intention space and the language space during prediction using the expression prediction model, including the range of similarity in the language space. In the example of FIG. 18, it is assumed that intention C in the intention space corresponds to "Yamakke" in the language space. Further, it is assumed that "Game Master" is included in the similar range of "Yamakke". At this time, the expression prediction model may predict "Yamakke" and "Gamer" in the linguistic space for the combination of feature amounts of intention C.
 図19は、言語空間における複数の埋め込み表現から、対応する重み係数を算出する際の言語空間と、意図空間の対応関係を模式的に示す図である。図19の例において、言語空間における「勝負師」と「逆張り」が選択されたとする。また、言語空間における「勝負師」と「逆張り」の中間点が、意図Cと意図Bからの距離の比がc:bであったとする。このとき、特徴量の重み係数は、例えば、(c÷(c+b))×(意図Cの重み係数)+(b÷(c+b))×(意図Bの重み係数)の式を用いて算出される。 FIG. 19 is a diagram schematically showing the correspondence between the language space and the intention space when calculating corresponding weighting coefficients from a plurality of embedded expressions in the language space. In the example of FIG. 19, it is assumed that "game player" and "contrarian" are selected in the language space. Further, suppose that the distance ratio between the midpoint between "game player" and "contrarian" in the linguistic space from intention C and intention B is c:b. At this time, the weighting coefficient of the feature amount is calculated using the formula: (c÷(c+b))×(weighting factor of intention C)+(b÷(c+b))×(weighting factor of intention B). Calculated using
 クエリ生成部46は、例えば、表現予測部45が予測した埋め込み表現に基づき、検索クエリを生成する。クエリ生成部46は、例えば、単語の埋め込み表現を検索する検索クエリを生成する。単語の埋め込み表現は、表現予測部45が表現予測モデルを用いて予測した重み係数の組み合わせに対応する。クエリ生成部46は、例えば、検索クエリとして、特徴量の重み係数の組み合わせに対応する単語の埋め込み表現を検索するプログラムを、SQLを用いて生成する。検索クエリは、SQL以外の言語を用いて検索クエリを生成してもよい。 The query generation unit 46 generates a search query based on the embedded expression predicted by the expression prediction unit 45, for example. For example, the query generation unit 46 generates a search query for searching for an embedded expression of a word. The embedded expression of a word corresponds to a combination of weighting coefficients predicted by the expression prediction unit 45 using an expression prediction model. For example, the query generation unit 46 generates, as a search query, a program using SQL to search for an embedded expression of a word corresponding to a combination of weighting coefficients of feature amounts. The search query may be generated using a language other than SQL.
 検索部47は、例えば、検索クエリに基づき、単語の埋め込み表現と同一または類似の表現が含まれる文章を検索する。検索部47は、例えば、ネットワーク上のデータから、クエリ生成部46が生成した検索クエリを用いて、検索クエリに対応する表現が含まれる文章を取得する。検索部47は、例えば、SNS(Social Networking Service)のデータから検索クエリに対応する表現が含まれる投稿を取得する。検索部47は、例えば、SNS以外のネットワーク上のデータから検索クエリに対応する表現が含まれる文章を取得してもよい。 For example, the search unit 47 searches for sentences that include expressions that are the same as or similar to the embedded expression of the word, based on the search query. For example, the search unit 47 uses the search query generated by the query generation unit 46 from data on the network to acquire a sentence that includes an expression corresponding to the search query. The search unit 47 obtains posts that include expressions corresponding to the search query from, for example, SNS (Social Networking Service) data. The search unit 47 may, for example, acquire a sentence that includes an expression corresponding to the search query from data on a network other than SNS.
 図20は、目的関数に含まれる特徴量の重み係数を対応する表現の検索結果をワードクラウドとして表示する表示画面の例を示す。図20の表示画面の例は、特徴量の重み係数の組み合わせに対応する表現と、類似する表現をワードクラウドとして表示している。図20の表示画面の例は、検索結果における、特徴量の重み係数の組み合わせに対応する表現と類似する表現の出現回数がワードクラウドの手法を用いて表示されている。 FIG. 20 shows an example of a display screen that displays search results for expressions corresponding to the weighting coefficients of the feature quantities included in the objective function as a word cloud. The example display screen in FIG. 20 displays expressions corresponding to combinations of weighting coefficients of feature amounts and similar expressions as a word cloud. In the example of the display screen in FIG. 20, the number of occurrences of expressions similar to expressions corresponding to the combination of weighting coefficients of feature amounts in the search results is displayed using a word cloud method.
 図21は、目的関数に含まれる特徴量の重み係数に対応する表現が含まれるSNSの投稿の検索結果の表示画面の例を示す。図21の表示画面の例では、特徴量の重み係数に対応する表現が含まれるSNSの投稿が表示されている。また、図21の表示画面の例では、下部に現在、用いている目的関数の意図に近い人の候補がAさんであることが表示されている。図21の表示画面の例で、投稿を見るボタンが押されると、検索部47は、例えば、SNSにおけるAさんの投稿を抽出する。そして、出力部41は、利用者端末装置20に、Aさんの投稿を出力する。 FIG. 21 shows an example of a display screen of search results of SNS posts that include expressions corresponding to weighting coefficients of feature quantities included in the objective function. In the example of the display screen in FIG. 21, an SNS post that includes an expression corresponding to the weighting coefficient of the feature amount is displayed. Further, in the example of the display screen in FIG. 21, it is displayed at the bottom that Mr. A is the candidate for the person who is close to the intention of the objective function currently being used. In the example of the display screen in FIG. 21, when the button to view posts is pressed, the search unit 47 extracts, for example, Mr. A's posts on the SNS. Then, the output unit 41 outputs Mr. A's post to the user terminal device 20.
 図22は、言語空間において、目的関数の更新に用いる意図を選択する表示画面の例を示す。図22の表示画面の例において、左側のグラフの「勝負師」と、「逆張り」が選択されたとする。図22の表示画面の例において、意図が選択された状態で、右に示す決定ボタンが押されると、更新部43は、係数予測モデルを用いて選択された意図に対応する特徴量の重み係数の組み合わせを予測する。そして、更新部43は、係数予測モデルが予測した特徴量の重み係数の組み合わせを用いて目的関数を更新する。 FIG. 22 shows an example of a display screen for selecting an intention to use for updating the objective function in the language space. In the example of the display screen in FIG. 22, it is assumed that "game player" and "contrarian" in the left graph are selected. In the example of the display screen in FIG. 22, when the decision button shown on the right is pressed with an intention selected, the update unit 43 uses the coefficient prediction model to update the weighting coefficient of the feature amount corresponding to the selected intention. Predict the combination of Then, the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction model.
 図23は、目的関数の更新に用いる意図を文章として入力する表示画面の例を示す。図23の表示画面の例において、左側の入力欄に意図を含む文章が入力される。図23の表示画面の例において、意図を含む文章が入力された状態で、右に示す決定ボタンが押されると、更新部43は、係数予測モデルを用いて、文章から意図に対応する表現を抽出し、抽出した表現に対応する特徴量の重み係数の組み合わせを予測する。そして、更新部43は、係数予測モデルが予測した特徴量の重み係数の組み合わせを用いて目的関数を更新する。 FIG. 23 shows an example of a display screen where the intention used to update the objective function is input as text. In the example of the display screen shown in FIG. 23, a sentence containing the intention is input into the input field on the left side. In the example of the display screen of FIG. 23, when the enter button shown on the right is pressed with a sentence containing an intention input, the update unit 43 uses the coefficient prediction model to extract an expression corresponding to the intention from the sentence. A combination of weighting coefficients of features corresponding to the extracted expressions is predicted. Then, the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction model.
 意思決定支援システム40が目的関数を更新する際の動作について説明する。図24は、意思決定支援システム40が目的関数を更新する際の動作フローの例を示す図である。 The operation when the decision support system 40 updates the objective function will be explained. FIG. 24 is a diagram illustrating an example of an operation flow when the decision support system 40 updates the objective function.
 出力部41は、更新対象となる目的関数に含まれる特徴量の重み係数を出力する(ステップS41)。出力部41は、例えば、利用者端末装置20に、更新対象となる目的関数に含まれる特徴量の重み係数を出力する。 The output unit 41 outputs the weighting coefficient of the feature quantity included in the objective function to be updated (step S41). The output unit 41 outputs, to the user terminal device 20, for example, the weighting coefficient of the feature amount included in the objective function to be updated.
 受付部42は、更新対象となる目的関数に含まれる特徴量の重み係数の変更値の入力を受け付ける(ステップS42)。受付部42は、例えば、利用者の操作によって利用者端末装置20に入力される、特徴量の重み係数の変更値を、利用者端末装置20から取得する。 The receiving unit 42 receives an input of a change value of a weighting coefficient of a feature included in the objective function to be updated (step S42). For example, the receiving unit 42 acquires from the user terminal device 20 a change value of the weighting coefficient of the feature quantity, which is input to the user terminal device 20 by a user's operation.
 特徴量の重み係数の変更値が取得されると、更新部43は、受付部42が取得した特徴量の重み係数の変更値を用いて目的関数を更新する(ステップS43)。目的関数を更新すると、更新部43は、更新した目的関数を保存する(ステップS44)。更新部43は、例えば、記憶部15に、更新した目的関数を保存する。更新された目的関数は、決定部12による最適解の決定に用いられる。 When the changed value of the weighting coefficient of the feature quantity is acquired, the updating unit 43 updates the objective function using the changed value of the weighting coefficient of the feature quantity acquired by the accepting unit 42 (step S43). After updating the objective function, the updating unit 43 saves the updated objective function (step S44). The updating unit 43 stores the updated objective function in the storage unit 15, for example. The updated objective function is used by the determining unit 12 to determine the optimal solution.
 意思決定支援システム40が単語の埋め込み表現を基に目的関数を更新する場合における動作について説明する。図25は、意思決定支援システム40が単語の埋め込み表現を基に目的関数を更新する際の動作フローの例を示す図である。 The operation when the decision support system 40 updates the objective function based on the embedded representation of words will be described. FIG. 25 is a diagram illustrating an example of an operation flow when the decision support system 40 updates the objective function based on the embedded expression of words.
 出力部41は、目的関数の更新に用いる、重み係数に対応する単語の埋め込み表現を出力する(ステップS51)。出力部41は、例えば、利用者端末装置20に、重み係数に対応する単語の埋め込み表現を出力する。 The output unit 41 outputs the embedded representation of the word corresponding to the weighting coefficient, which is used to update the objective function (step S51). The output unit 41 outputs, for example, the embedded representation of the word corresponding to the weighting coefficient to the user terminal device 20.
 受付部42は、変更後の埋め込み表現の入力を受け付ける(ステップS52)。受付部42は、例えば、利用者の操作によって利用者端末装置20に入力される、変更後の埋め込み表現を、利用者端末装置20から取得する。 The accepting unit 42 accepts input of the modified embedded expression (step S52). For example, the reception unit 42 acquires from the user terminal device 20 a modified embedded expression that is input to the user terminal device 20 by a user's operation.
 変更後の埋め込み表現が取得されると、係数予測部44は、係数予測モデルを用いて、取得された埋め込み表現に対応する特徴量の重み係数の組み合わせを予測する(ステップS53)。特徴量の重み係数の組み合わせが予測されると、更新部43は、係数予測部44が予測した特徴量の重み係数の組み合わせを用いて目的関数を更新する(ステップS54)。目的関数を更新すると、更新部43は、更新した目的関数を保存する(ステップS55)。更新部43は、例えば、記憶部15に、更新した目的関数を保存する更新された目的関数は、決定部12による最適解の決定に用いられる。 When the modified embedded expression is acquired, the coefficient prediction unit 44 uses the coefficient prediction model to predict the combination of weighting coefficients of the feature amounts corresponding to the acquired embedded expression (step S53). When the combination of weighting coefficients of the feature quantities is predicted, the updating unit 43 updates the objective function using the combination of weighting coefficients of the feature quantities predicted by the coefficient prediction unit 44 (step S54). After updating the objective function, the updating unit 43 saves the updated objective function (step S55). The updating unit 43 stores the updated objective function in, for example, the storage unit 15. The updated objective function is used by the determining unit 12 to determine the optimal solution.
 意思決定支援システム40が目的関数に含まれる特徴量の重み係数に対応する表現が含まれる文章を抽出する際の動作について説明する。図26は、意思決定支援システム40が目的関数に含まれる特徴量の重み係数に対応する表現を含む文章を抽出する際の動作フローの例を示す図である。 The operation when the decision support system 40 extracts a sentence that includes an expression corresponding to the weighting coefficient of the feature included in the objective function will be described. FIG. 26 is a diagram illustrating an example of an operation flow when the decision support system 40 extracts a sentence that includes an expression that corresponds to the weighting coefficient of the feature included in the objective function.
 表現予測部45は、表現予測モデルを用いて、最適解の決定に用いられる目的関数に含まれる特徴量の重み係数に対応する表現を予測する(ステップS61)。表現予測部45は、例えば、目的関数に含まれる特徴量の重み係数から、言語空間上の表現を予測する表現予測モデルを用いて、最適解の決定に用いられる目的関数に含まれる特徴量の重み係数に対応する単語の埋め込み表現を予測する。対応する表現が予測されると、クエリ生成部46は、対応する表現が含まれる文章を検索する検索クエリを生成する(ステップS62)。 The expression prediction unit 45 uses the expression prediction model to predict an expression corresponding to the weighting coefficient of the feature included in the objective function used to determine the optimal solution (step S61). The expression prediction unit 45 uses, for example, an expression prediction model that predicts an expression in language space from the weighting coefficients of the features included in the objective function, and calculates the features included in the objective function used to determine the optimal solution. Predict the embedded representation of the word corresponding to the weighting coefficient. When the corresponding expression is predicted, the query generation unit 46 generates a search query to search for sentences that include the corresponding expression (step S62).
 検索クエリが生成されると、検索部47は、例えば、クエリ生成部46が生成した検索クエリを用いて、ネットワーク上のデータから検索クエリに対応する文章を検索する(ステップS63)。検索部47は、例えば、ネットワーク上のSNSのデータから検索クエリに対応する投稿を取得する。 Once the search query is generated, the search unit 47 uses, for example, the search query generated by the query generation unit 46 to search the data on the network for a sentence corresponding to the search query (step S63). The search unit 47 obtains posts corresponding to the search query from, for example, SNS data on the network.
 検索クエリに対応する表現を含む文章が取得されると、出力部41は、検索クエリを用いた検索結果を出力する(ステップS64)。出力部41は、例えば、利用者端末装置20に、検索クエリを用いた検索結果を出力する。 When a sentence including an expression corresponding to the search query is acquired, the output unit 41 outputs a search result using the search query (step S64). The output unit 41 outputs a search result using a search query to the user terminal device 20, for example.
 意思決定支援システム40は、目的関数に含まれる特徴量の重み係数の変更値を取得する。そして、意思決定支援システム40は、変更値を用いて目的関数を更新する。更新した目的関数を用いることで、例えば、レース結果の予測に関する最適解を利用する人物に応じて最適化された最適解の決定を行うことができる。 The decision support system 40 obtains the change value of the weighting coefficient of the feature quantity included in the objective function. The decision support system 40 then updates the objective function using the changed value. By using the updated objective function, for example, it is possible to determine an optimal solution that is optimized depending on the person who will use the optimal solution regarding prediction of race results.
 目的関数に含まれる特徴量の重み係数をグラフ上に表示して変更値の入力を受け付ける場合に、意思決定支援システム40は、グラフ上の重み係数の表示状態の変更結果を基に、目的関数を更新する。このような方法で目的関数を更新することで、目的関数の最適化が容易になる。目的関数の最適化を行うことで、レース結果の予測における最適解を利用する人物により適した最適解を得ることができる。 When displaying the weighting coefficients of the feature quantities included in the objective function on a graph and accepting input of change values, the decision support system 40 changes the objective function based on the result of changing the display state of the weighting coefficients on the graph. Update. By updating the objective function in this way, optimization of the objective function becomes easier. By optimizing the objective function, it is possible to obtain an optimal solution that is more suitable for the person who uses the optimal solution in predicting the race result.
 また、目的関数に対応する意図の選択結果を基に目的関数を更新する場合には、利用者が特徴量の重み係数の変更に関する知識がなくても容易に目的関数を更新することができる。 Furthermore, when updating the objective function based on the selection result of the intention corresponding to the objective function, the user can easily update the objective function even if the user does not have knowledge about changing the weighting coefficient of the feature amount.
 また、意思決定支援システム40は、目的関数に含まれる特徴量の重み係数に対応する、言語空間上の表現を含む文章を抽出する。そして、意思決定支援システム40は、例えば、利用者端末装置20に、抽出した文章を出力する。このような処理を行うことで、意思決定支援システム40は、最適解の決定理由に関係する情報を出力することができる。この結果、意思決定支援システム40の利用者は、最適解の決定理由に関係する情報を参照して、最適解の決定結果を解釈することができる。 Furthermore, the decision support system 40 extracts sentences that include expressions in the linguistic space that correspond to the weighting coefficients of the features included in the objective function. The decision support system 40 then outputs the extracted text to the user terminal device 20, for example. By performing such processing, the decision support system 40 can output information related to the reason for determining the optimal solution. As a result, the user of the decision support system 40 can refer to information related to the reason for determining the optimal solution and interpret the result of determining the optimal solution.
 (第3の実施形態)
 本発明の第3の実施形態について図を参照して詳細に説明する。図27は、本実施形態の意思決定支援システム100の構成の例を示す図である。意思決定支援システム100は、取得部101と、決定部102と、出力部103を備える。
(Third embodiment)
A third embodiment of the present invention will be described in detail with reference to the drawings. FIG. 27 is a diagram showing an example of the configuration of the decision support system 100 of this embodiment. The decision support system 100 includes an acquisition section 101, a determination section 102, and an output section 103.
 取得部101は、公営競技のレースに関する情報を取得する。決定部102は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する。出力部103は、決定した最適解を基に、レース結果の予測に関する情報を出力する。 The acquisition unit 101 acquires information regarding races in publicly managed competitions. The determining unit 102 uses an objective function generated in advance by inverse reinforcement learning based on the decision history regarding the prediction of race results in publicly managed competitions, and the obtained information regarding races in publicly managed competitions to determine the optimal solution for predicting race results. Determine. The output unit 103 outputs information regarding prediction of the race result based on the determined optimal solution.
 ここで、第1の実施形態の取得部11は、取得部101の一例である。また、取得部101は、取得手段の一態様である。第1の実施形態の決定部12は、決定部102の一例である。また、決定部102は、決定手段の一態様である。第1の実施形態の出力部13は、出力部103の一例である。また、出力部103は、出力手段の一態様である。 Here, the acquisition unit 11 of the first embodiment is an example of the acquisition unit 101. Further, the acquisition unit 101 is one aspect of an acquisition means. The determining unit 12 of the first embodiment is an example of the determining unit 102. Furthermore, the determining unit 102 is one aspect of determining means. The output unit 13 of the first embodiment is an example of the output unit 103. Furthermore, the output unit 103 is one aspect of output means.
 意思決定支援システム100の動作について説明する。図28は、意思決定支援システム100の動作フローの例を示す図である。 The operation of the decision support system 100 will be explained. FIG. 28 is a diagram illustrating an example of the operation flow of the decision support system 100.
 取得部101は、公営競技のレースに関する情報を取得する(ステップS101)。公営競技のレースに関する情報が取得されると、決定部102は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する(ステップS102)。最適解が決定されると、出力部103は、決定した最適解を基に、レース結果の予測に関する情報を出力する(ステップS103)。 The acquisition unit 101 acquires information regarding races in publicly managed competitions (step S101). When the information regarding the race in the publicly managed competition is acquired, the determining unit 102 uses an objective function generated in advance by inverse reinforcement learning based on the decision history regarding prediction of the race result in the publicly managed competition and the acquired information regarding the race in the publicly managed competition. The optimal solution for predicting the race result is determined using the following (step S102). Once the optimal solution is determined, the output unit 103 outputs information regarding prediction of the race result based on the determined optimal solution (step S103).
 本実施形態の意思決定支援システム100は、公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、公営競技のレースに関する情報とを用いて、レース結果の予測のおける最適解を決定する。そして、意思決定支援システム100は、決定した最適解を出力する。この結果、意思決定支援システム100を用いることで、公営競技のレース結果に関する意思決定を容易に行うことができる。 The decision-making support system 100 of this embodiment uses an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding the prediction of race results in publicly managed competitions, and information regarding races in publicly managed competitions. Determine the optimal solution for prediction. The decision support system 100 then outputs the determined optimal solution. As a result, by using the decision-making support system 100, it is possible to easily make decisions regarding the race results of publicly managed competitions.
 第1の実施形態の意思決定支援システム10、第2の実施形態の意思決定支援システム40および第2の実施形態の意思決定支援システム100における各処理は、コンピュータプログラムをコンピュータで実行することによって実現することができる。図29は、第1の実施形態の意思決定支援システム10、第2の実施形態の意思決定支援システム40および第3の実施形態の意思決定支援システム100における各処理を行うコンピュータプログラムを実行するコンピュータ200の構成の例を示したものである。コンピュータ200は、CPU(Central Processing Unit)201と、メモリ202と、記憶装置203と、入出力I/F(Interface)204と、通信I/F205を備える。 Each process in the decision support system 10 of the first embodiment, the decision support system 40 of the second embodiment, and the decision support system 100 of the second embodiment is realized by executing a computer program on a computer. can do. FIG. 29 shows a computer that executes a computer program that performs each process in the decision support system 10 of the first embodiment, the decision support system 40 of the second embodiment, and the decision support system 100 of the third embodiment. 200 shows an example of a configuration of 200. The computer 200 includes a CPU (Central Processing Unit) 201, a memory 202, a storage device 203, an input/output I/F (Interface) 204, and a communication I/F 205.
 CPU201は、記憶装置203から各処理を行うコンピュータプログラムを読み出して実行する。CPU201は、複数のCPUの組み合わせによって構成されていてもよい。また、CPU201は、CPUと他の種類のプロセッサの組み合わせによって構成されていてもよい。例えば、CPU201は、CPUとGPU(Graphics Processing Unit)の組み合わせによって構成されていてもよい。メモリ202は、DRAM(Dynamic Random Access Memory)等によって構成され、CPU201が実行するコンピュータプログラムや処理中のデータが一時記憶される。記憶装置203は、CPU201が実行するコンピュータプログラムを記憶している。記憶装置203は、例えば、不揮発性の半導体記憶装置によって構成されている。記憶装置203には、ハードディスクドライブ等の他の記憶装置が用いられてもよい。入出力I/F204は、作業者からの入力の受付および表示データ等の出力を行うインタフェースである。通信I/F205は、利用者端末装置20および情報管理サーバ30との間でデータの送受信を行うインタフェースである。また、利用者端末装置20および情報管理サーバ30も同様の構成としてもよい。 The CPU 201 reads computer programs for performing each process from the storage device 203 and executes them. The CPU 201 may be configured by a combination of multiple CPUs. Further, the CPU 201 may be configured by a combination of a CPU and other types of processors. For example, the CPU 201 may be configured by a combination of a CPU and a GPU (Graphics Processing Unit). The memory 202 is configured with a DRAM (Dynamic Random Access Memory) or the like, and temporarily stores computer programs executed by the CPU 201 and data being processed. The storage device 203 stores computer programs executed by the CPU 201. The storage device 203 is configured by, for example, a nonvolatile semiconductor storage device. Other storage devices such as a hard disk drive may be used as the storage device 203. The input/output I/F 204 is an interface that receives input from a worker and outputs display data and the like. The communication I/F 205 is an interface that transmits and receives data between the user terminal device 20 and the information management server 30. Further, the user terminal device 20 and the information management server 30 may also have similar configurations.
 各処理の実行に用いられるコンピュータプログラムは、データを非一時的に記録するコンピュータ読み取り可能な記録媒体に格納して頒布することもできる。記録媒体としては、例えば、データ記録用磁気テープや、ハードディスクなどの磁気ディスクを用いることができる。また、記録媒体としては、CD-ROM(Compact Disc Read Only Memory)等の光ディスクを用いることもできる。不揮発性の半導体記憶装置を記録媒体として用いてもよい。 The computer program used to execute each process can also be stored and distributed in a computer-readable recording medium that non-temporarily records data. As the recording medium, for example, a magnetic tape for data recording or a magnetic disk such as a hard disk can be used. Further, as the recording medium, an optical disc such as a CD-ROM (Compact Disc Read Only Memory) can also be used. A nonvolatile semiconductor memory device may be used as the recording medium.
 上記の実施形態の一部又は全部は、以下の付記のようにも記載されうるが、以下には限られない。
[付記1]
 公営競技のレースに関する情報を取得する取得手段と、
 公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する決定手段と、
 決定した最適解を基に、レース結果の予測に関する情報を出力する出力手段と
 を備える意思決定支援システム。
[付記2]
 前記レース結果の予測における最適解は、購入する投票券の組み合わせである、
 付記1に記載の意思決定支援システム。
[付記3]
 前記出力手段は、前記目的関数に含まれる特徴量の重み係数の組み合わせを可視化したデータを出力する、
 付記1または2に記載の意思決定支援システム。
[付記4]
 前記目的関数に含まれる特徴量の重み係数を変更する変更値を受け付ける受付手段と、
 受け付けた変更値を用いて前記目的関数を更新する更新手段と
 をさらに備え、
 前記決定手段は、更新した前記目的関数を用いて、前記最適解を決定する、
 付記1から3いずれかに記載の意思決定支援システム。
[付記5]
 前記出力手段は、前記目的関数に含まれる特徴量の重み係数をグラフとして出力し、
 前記受付手段は、前記グラフ上に表示される前記目的関数に含まれる特徴量の重み係数を変更することで入力される前記変更値を受け付ける、
 付記4に記載の意思決定支援システム。
[付記6]
 単語の埋め込み表現から前記目的関数に含まれる特徴量の重み係数の組み合わせを予測する係数予測モデルを用いて、前記単語の埋め込み表現から、前記重み係数の組み合わせを予測する係数予測手段と、
 前記係数予測手段が予測した前記重み係数の組み合わせを基に、前記目的関数を更新する更新手段と、をさらに備える、
 付記1から3いずれかに記載の意思決定支援システム。
[付記7]
 前記目的関数に含まれる特徴量の重み係数の組み合わせから単語の埋め込み表現を予測する表現予測モデルを用いて、前記目的関数に含まれる特徴量の重み係数の組み合わせから、前記重み係数の組み合わせに対応する埋め込み表現を予測する表現予測手段をさらに備える、
 付記1から6いずれかに記載の意思決定支援システム。
[付記8]
 前記出力手段は、前記表現予測手段が予測した埋め込み表現を可視化したデータを出力する、
 付記7に記載の意思決定支援システム。
[付記9]
 前記表現予測手段が予測した埋め込み表現に基づき、検索クエリを生成するクエリ生成手段をさらに備える、
 付記7または8に記載の意思決定支援システム。
[付記10]
 前記検索クエリに基づき、前記埋め込み表現と同一または類似の表現が含まれる文章を検索する検索手段をさらに備える、
 付記9に記載の意思決定支援システム。
[付記11]
 公営競技のレースに関する情報からレース結果の予測における最適解を決定する目的関数を、公営競技のレース結果の予測に関する意思決定履歴に基づいた逆強化学習によって生成する生成手段をさらに備える、
 付記1から10いずれかに記載の意思決定支援システム。
[付記12]
 前記公営競技は、競馬であり、
 公営競技のレースに関する情報は、レース条件と、出走馬の属性と、オッズに関するデータを含む、
 付記1から11いずれかに記載の意思決定支援システム。
[付記13]
 公営競技のレースに関する情報を取得し、
 公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定し、
 決定した最適解を基に、レース結果の予測に関する情報を出力する、
 意思決定支援方法。
[付記14]
 公営競技のレースに関する情報を取得する処理と、
 公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する処理と、
 決定した最適解を基に、レース結果の予測に関する情報を出力する処理と、
 をコンピュータに実行させる意思決定支援プログラムを非一時的に記録する記録媒体。
Part or all of the above embodiments may be described as in the following additional notes, but are not limited to the following.
[Additional note 1]
an acquisition means for acquiring information regarding races in publicly managed competitions;
A determining means that determines an optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on decision-making history regarding prediction of race results in publicly managed competitions and acquired information regarding races in publicly managed competitions. and,
A decision support system comprising: an output means for outputting information regarding prediction of a race result based on the determined optimal solution.
[Additional note 2]
The optimal solution in predicting the race result is a combination of voting tickets to purchase,
Decision support system described in Appendix 1.
[Additional note 3]
The output means outputs data visualizing combinations of weighting coefficients of features included in the objective function.
Decision support system according to appendix 1 or 2.
[Additional note 4]
reception means for accepting a change value for changing a weighting coefficient of a feature included in the objective function;
and updating means for updating the objective function using the received change value,
The determining means determines the optimal solution using the updated objective function.
The decision support system described in any of Supplementary Notes 1 to 3.
[Additional note 5]
The output means outputs weighting coefficients of features included in the objective function as a graph,
The receiving means receives the changed value input by changing a weighting coefficient of a feature included in the objective function displayed on the graph.
Decision support system described in Appendix 4.
[Additional note 6]
Coefficient prediction means for predicting the combination of weighting coefficients from the embedded representation of the word using a coefficient prediction model that predicts the combination of weighting coefficients of the feature quantities included in the objective function from the embedded representation of the word;
further comprising: updating means for updating the objective function based on the combination of the weighting coefficients predicted by the coefficient predicting means;
The decision support system described in any of Supplementary Notes 1 to 3.
[Additional note 7]
Using an expression prediction model that predicts an embedded expression of a word from a combination of weighting coefficients of features included in the objective function, a combination of weighting coefficients is calculated from a combination of weighting coefficients of features included in the objective function. further comprising expression prediction means for predicting an embedded expression for
The decision support system according to any one of Supplementary Notes 1 to 6.
[Additional note 8]
The output means outputs data that visualizes the embedded expression predicted by the expression prediction means.
Decision support system described in Appendix 7.
[Additional note 9]
further comprising query generation means for generating a search query based on the embedded expression predicted by the expression prediction means;
Decision support system according to appendix 7 or 8.
[Additional note 10]
further comprising a search means for searching for sentences containing expressions that are the same as or similar to the embedded expression, based on the search query;
Decision support system described in Appendix 9.
[Additional note 11]
Further comprising a generation means for generating an objective function for determining an optimal solution for predicting race results from information regarding races in publicly managed competitions by inverse reinforcement learning based on decision-making history regarding prediction of race results in publicly managed competitions;
The decision support system according to any one of Supplementary Notes 1 to 10.
[Additional note 12]
The publicly managed competition is horse racing;
Information regarding races in publicly managed competitions includes data on race conditions, attributes of participating horses, and odds.
The decision support system according to any one of Supplementary Notes 1 to 11.
[Additional note 13]
Obtain information about publicly managed races,
Determine the optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in public competitions and the obtained information regarding races in public competitions,
Outputs information regarding prediction of race results based on the determined optimal solution.
Decision support methods.
[Additional note 14]
Processing to obtain information regarding races in publicly managed competitions;
A process of determining an optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on decision-making history regarding prediction of race results in public competitions and acquired information regarding races in public competitions. ,
A process of outputting information regarding the prediction of the race result based on the determined optimal solution;
A recording medium that non-temporarily records a decision support program that causes a computer to execute.
 以上、上述した実施形態を例として本発明を説明した。しかしながら、本発明は、上述した実施形態には限定されない。即ち、本発明は、本発明のスコープ内において、当業者が理解し得る様々な態様を適用することができる。 The present invention has been described above using the above-described embodiment as an example. However, the invention is not limited to the embodiments described above. That is, the present invention can apply various aspects that can be understood by those skilled in the art within the scope of the present invention.
 10  意思決定支援システム
 11  取得部
 12  決定部
 13  出力部
 14  生成部
 15  記憶部
 20  利用者端末装置
 30  情報管理サーバ
 40  意思決定支援システム
 41  出力部
 42  受付部
 43  更新部
 44  係数予測部
 45  表現予測部
 46  クエリ生成部
 47  検索部
 100  意思決定支援システム
 101  取得部
 102  決定部
 103  出力部
 200  コンピュータ
 201  CPU
 202  メモリ
 203  記憶装置
 204  入出力I/F
 205  通信I/F
10 Decision Support System 11 Acquisition Unit 12 Determination Unit 13 Output Unit 14 Generation Unit 15 Storage Unit 20 User Terminal Device 30 Information Management Server 40 Decision Support System 41 Output Unit 42 Reception Unit 43 Update Unit 44 Coefficient Prediction Unit 45 Expression Prediction Section 46 Query generation section 47 Search section 100 Decision support system 101 Acquisition section 102 Determination section 103 Output section 200 Computer 201 CPU
202 Memory 203 Storage device 204 Input/output I/F
205 Communication I/F

Claims (14)

  1.  公営競技のレースに関する情報を取得する取得手段と、
     公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する決定手段と、
     決定した最適解を基に、レース結果の予測に関する情報を出力する出力手段と
     を備える意思決定支援システム。
    an acquisition means for acquiring information regarding races in publicly managed competitions;
    A determining means that determines an optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on decision-making history regarding prediction of race results in publicly managed competitions and acquired information regarding races in publicly managed competitions. and,
    A decision support system comprising: an output means for outputting information regarding prediction of a race result based on the determined optimal solution.
  2.  前記レース結果の予測における最適解は、購入するレースの投票券の組み合わせである、
     請求項1に記載の意思決定支援システム。
    The optimal solution in predicting the race result is a combination of voting tickets for the races to be purchased.
    The decision support system according to claim 1.
  3.  前記出力手段は、前記目的関数に含まれる特徴量の重み係数の組み合わせを可視化したデータを出力する、
     請求項1または2に記載の意思決定支援システム。
    The output means outputs data visualizing combinations of weighting coefficients of features included in the objective function.
    The decision support system according to claim 1 or 2.
  4.  前記目的関数に含まれる特徴量の重み係数を変更する変更値を受け付ける受付手段と、
     受け付けた前記変更値を用いて前記目的関数を更新する更新手段と
     をさらに備え、
     前記決定手段は、更新した前記目的関数を用いて、前記最適解を決定する、
     請求項1から3いずれかに記載の意思決定支援システム。
    reception means for accepting a change value for changing a weighting coefficient of a feature included in the objective function;
    and updating means for updating the objective function using the received changed value,
    The determining means determines the optimal solution using the updated objective function.
    A decision support system according to any one of claims 1 to 3.
  5.  前記出力手段は、前記目的関数に含まれる特徴量の重み係数をグラフとして出力し、
     前記受付手段は、前記グラフ上に表示される前記目的関数に含まれる特徴量の重み係数を変更することで入力される前記変更値を受け付ける、
     請求項4に記載の意思決定支援システム。
    The output means outputs weighting coefficients of features included in the objective function as a graph,
    The receiving means receives the changed value input by changing a weighting coefficient of a feature included in the objective function displayed on the graph.
    The decision support system according to claim 4.
  6.  単語の埋め込み表現から前記目的関数に含まれる特徴量の重み係数の組み合わせを予測する係数予測モデルを用いて、前記単語の埋め込み表現から、前記重み係数の組み合わせを予測する係数予測手段と、
     前記係数予測手段が予測した前記重み係数の組み合わせを基に、前記目的関数を更新する更新手段と、をさらに備える、
     請求項1から3いずれかに記載の意思決定支援システム。
    Coefficient prediction means for predicting the combination of weighting coefficients from the embedded representation of the word using a coefficient prediction model that predicts the combination of weighting coefficients of the feature quantities included in the objective function from the embedded representation of the word;
    further comprising: updating means for updating the objective function based on the combination of the weighting coefficients predicted by the coefficient predicting means;
    A decision support system according to any one of claims 1 to 3.
  7.  前記目的関数に含まれる特徴量の重み係数の組み合わせから単語の埋め込み表現を予測する表現予測モデルを用いて、前記目的関数に含まれる特徴量の重み係数の組み合わせから、前記重み係数の組み合わせに対応する埋め込み表現を予測する表現予測手段をさらに備える、
     請求項1から6いずれかに記載の意思決定支援システム。
    Using an expression prediction model that predicts an embedded expression of a word from a combination of weighting coefficients of features included in the objective function, a combination of weighting coefficients is calculated from a combination of weighting coefficients of features included in the objective function. further comprising expression prediction means for predicting an embedded expression for
    A decision support system according to any one of claims 1 to 6.
  8.  前記出力手段は、前記表現予測手段が予測した埋め込み表現を可視化したデータを出力する、
     請求項7に記載の意思決定支援システム。
    The output means outputs data that visualizes the embedded expression predicted by the expression prediction means.
    The decision support system according to claim 7.
  9.  前記表現予測手段が予測した埋め込み表現に基づき、検索クエリを生成するクエリ生成手段をさらに備える、
     請求項7または8に記載の意思決定支援システム。
    further comprising query generation means for generating a search query based on the embedded expression predicted by the expression prediction means;
    The decision support system according to claim 7 or 8.
  10.  前記検索クエリに基づき、前記埋め込み表現と同一または類似の表現が含まれる文章を検索する検索手段をさらに備える、
     請求項9に記載の意思決定支援システム。
    further comprising a search means for searching for sentences containing expressions that are the same as or similar to the embedded expression, based on the search query;
    The decision support system according to claim 9.
  11.  公営競技のレースに関する情報からレース結果の予測における最適解を決定する目的関数を、公営競技のレース結果の予測に関する意思決定履歴に基づいた逆強化学習によって生成する生成手段をさらに備える、
     請求項1から10いずれかに記載の意思決定支援システム。
    Further comprising a generation means for generating an objective function for determining an optimal solution for predicting race results from information regarding races in publicly managed competitions by inverse reinforcement learning based on decision-making history regarding prediction of race results in publicly managed competitions;
    A decision support system according to any one of claims 1 to 10.
  12.  前記公営競技は、競馬であり、
     公営競技のレースに関する情報は、レース条件と、出走馬の属性と、オッズに関するデータを含む、
     請求項1から11いずれかに記載の意思決定支援システム。
    The publicly managed competition is horse racing;
    Information regarding races in publicly managed competitions includes data on race conditions, attributes of participating horses, and odds.
    A decision support system according to any one of claims 1 to 11.
  13.  公営競技のレースに関する情報を取得し、
     公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定し、
     決定した最適解を基に、レース結果の予測に関する情報を出力する、
     意思決定支援方法。
    Obtain information about publicly managed races,
    Determine the optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on the decision-making history regarding prediction of race results in public competitions and the obtained information regarding races in public competitions,
    Outputs information regarding prediction of race results based on the determined optimal solution.
    Decision support methods.
  14.  公営競技のレースに関する情報を取得する処理と、
     公営競技のレース結果の予測に関する意思決定履歴に基づき逆強化学習で予め生成された目的関数と、取得した公営競技のレースに関する情報とを用いて、レース結果の予測における最適解を決定する処理と、
     決定した最適解を基に、レース結果の予測に関する情報を出力する処理と、
     をコンピュータに実行させる意思決定支援プログラムを非一時的に記録する記録媒体。
    Processing to obtain information regarding races in publicly managed competitions;
    A process of determining an optimal solution for predicting race results using an objective function generated in advance by inverse reinforcement learning based on decision-making history regarding prediction of race results in public competitions and acquired information regarding races in public competitions. ,
    A process of outputting information regarding the prediction of the race result based on the determined optimal solution;
    A recording medium that non-temporarily records a decision support program that causes a computer to execute.
PCT/JP2022/012639 2022-03-18 2022-03-18 Decision support system, decision support method, and recording medium WO2023175910A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2022/012639 WO2023175910A1 (en) 2022-03-18 2022-03-18 Decision support system, decision support method, and recording medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2022/012639 WO2023175910A1 (en) 2022-03-18 2022-03-18 Decision support system, decision support method, and recording medium

Publications (1)

Publication Number Publication Date
WO2023175910A1 true WO2023175910A1 (en) 2023-09-21

Family

ID=88022635

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2022/012639 WO2023175910A1 (en) 2022-03-18 2022-03-18 Decision support system, decision support method, and recording medium

Country Status (1)

Country Link
WO (1) WO2023175910A1 (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101976988B1 (en) * 2018-06-21 2019-08-28 주식회사 미디어피아 Server and method for forecasting the probability of horse racing
JP2020149601A (en) * 2019-03-15 2020-09-17 エヌ・ティ・ティ・コミュニケーションズ株式会社 Data processing device, data processing method and data processing program
WO2021229626A1 (en) * 2020-05-11 2021-11-18 日本電気株式会社 Learning device, learning method, and learning program

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101976988B1 (en) * 2018-06-21 2019-08-28 주식회사 미디어피아 Server and method for forecasting the probability of horse racing
JP2020149601A (en) * 2019-03-15 2020-09-17 エヌ・ティ・ティ・コミュニケーションズ株式会社 Data processing device, data processing method and data processing program
WO2021229626A1 (en) * 2020-05-11 2021-11-18 日本電気株式会社 Learning device, learning method, and learning program

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
TAKUMI YOSHIDA, YOKOYAMA SOICHIRO, YAMASHITA TOMOHISA, KAWAMURA HIDENOR: "Automatic Content Generation for Purchasing Support of Bicycle Tickets", 119TH KNOWLEDGE BASE SYSTEM STUDY GROUP MATERIALS (SIG-KBS-B903), 1 March 2020 (2020-03-01) - 10 March 2020 (2020-03-10), pages 17 - 22, XP093092599 *

Similar Documents

Publication Publication Date Title
WO2007065125A2 (en) Method for selecting media products
Xu et al. Entrepreneurial learning and disincentives in crowdfunding markets
JP2015127922A (en) Information processing device, information processing method, information processing program and information processing system
JP5702422B2 (en) Information processing apparatus, information processing method, and information processing program
Paravizo et al. Playing for real: an exploratory analysis of professional esports athletes’ work
Patnaik et al. A study of Prediction models for football player valuations by quantifying statistical and economic attributes for the global transfer market
WO2023175910A1 (en) Decision support system, decision support method, and recording medium
Scarf et al. On skill and chance in sport
KR102223640B1 (en) Cloud-based personalized contents subscription service providing system and method thereof
Ekstrøm et al. Evaluating one-shot tournament predictions
CN110363574A (en) Products Show method, apparatus, electronic equipment and readable medium
CN115738285A (en) Game quality evaluation feedback method and system
JP2020149583A (en) Program, method, and system for predicting potential ability of race horse, seed horse candidate presentation program, method for presenting seed horse candidate, seed horse candidate presentation system, and prediction career earnings database preparation program used for the same
WO2023170832A1 (en) Prediction system, prediction method, and recording medium
Lin et al. Research on User Experience Optimization of Tutorial Design for Battle Royale Games Based on Grey AHP Theory
WO2023175809A1 (en) Situation prediction system, situation prediction method, and recording medium
Powell Generalizing the Elo rating system for multiplayer games and races: why endurance is better than speed
Lau-Soto An agent-based model of a betting exchange via a single-threaded and object-oriented implementation
JP5509126B2 (en) Predictive power diagnosis apparatus and predictive power diagnosis method
JP5839765B1 (en) Information processing apparatus, information processing method, and information processing program
De Pessemier et al. Recommendations for sports games to bet on
WO2023053882A1 (en) Betting ticket purchasing support system, betting ticket purchasing support method, and betting ticket purchasing support program
Li et al. A social selection mechanism for sports betting market
Mahmood et al. Psl eye: Predicting the winning team in pakistan super league (psl) matches
Malagón-Selma et al. Measuring the popularity of football players with Google Trends

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22932183

Country of ref document: EP

Kind code of ref document: A1