WO2026016754A1 - 一种目标对象的确定方法、装置、设备及存储介质 - Google Patents

一种目标对象的确定方法、装置、设备及存储介质

Info

Publication number
WO2026016754A1
WO2026016754A1 PCT/CN2025/103408 CN2025103408W WO2026016754A1 WO 2026016754 A1 WO2026016754 A1 WO 2026016754A1 CN 2025103408 W CN2025103408 W CN 2025103408W WO 2026016754 A1 WO2026016754 A1 WO 2026016754A1
Authority
WO
WIPO (PCT)
Prior art keywords
match
result
contribution
participant
prediction model
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
PCT/CN2025/103408
Other languages
English (en)
French (fr)
Inventor
熊宇
孙海峰
吴润泽
王凯
吕唐杰
范长杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Netease Hangzhou Network Co Ltd
Original Assignee
Netease Hangzhou Network Co Ltd
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 Netease Hangzhou Network Co Ltd filed Critical Netease Hangzhou Network Co Ltd
Publication of WO2026016754A1 publication Critical patent/WO2026016754A1/zh
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • AHUMAN NECESSITIES
    • A63SPORTS; GAMES; AMUSEMENTS
    • A63FCARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/70Game security or game management aspects
    • A63F13/79Game security or game management aspects involving player-related data, e.g. identities, accounts, preferences or play histories
    • A63F13/798Game security or game management aspects involving player-related data, e.g. identities, accounts, preferences or play histories for assessing skills or for ranking players, e.g. for generating a hall of fame
    • AHUMAN NECESSITIES
    • A63SPORTS; GAMES; AMUSEMENTS
    • A63FCARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F2300/00Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game
    • A63F2300/50Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game characterized by details of game servers
    • A63F2300/55Details of game data or player data management
    • A63F2300/5546Details of game data or player data management using player registration data, e.g. identification, account, preferences, game history
    • A63F2300/556Player lists, e.g. online players, buddy list, black list
    • AHUMAN NECESSITIES
    • A63SPORTS; GAMES; AMUSEMENTS
    • A63FCARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F2300/00Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game
    • A63F2300/50Features of games using an electronically generated display having two or more dimensions, e.g. on a television screen, showing representations related to the game characterized by details of game servers
    • A63F2300/55Details of game data or player data management
    • A63F2300/5546Details of game data or player data management using player registration data, e.g. identification, account, preferences, game history
    • A63F2300/558Details of game data or player data management using player registration data, e.g. identification, account, preferences, game history by assessing the players' skills or ranking

Definitions

  • This disclosure relates to the field of data processing technology, and more specifically, to a method, apparatus, device, and storage medium for determining a target object.
  • Existing target selection schemes can be divided into two categories: The first category is to obtain the performance score of each participant in the competition by weighted summing based on the various performance characteristics of the participants in the competition and the weight coefficient corresponding to each performance characteristic, and select the participant with the highest performance score as the best performing target in the competition (mostly used in online competitions); The second category is to determine the participant with the most expert votes as the best performing target in the competition (mostly used in offline competitions).
  • the two target selection schemes mentioned above rely on human experience in terms of the weighting coefficients corresponding to different performance characteristics and the specific voting by experts. As a result, the final target selection results cannot be explained using more objective data, which makes the target selection results susceptible to interference from human factors and have poor interpretability.
  • the present disclosure provides a method, apparatus, device and storage medium for determining target objects.
  • it provides a more fair, objective and reasonable basis for the selection of target objects, making the target object selection results simple, transparent and highly interpretable. This is conducive to helping various competition organizers to more comprehensively and objectively evaluate the performance of each participant.
  • embodiments of this disclosure provide a method for determining a target object, the method comprising:
  • the contribution of the performance characteristics to the match prediction results output by the match prediction model is determined, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the target of this competition is determined from the participants.
  • embodiments of this disclosure provide a target object determination device, the determination device comprising:
  • the contribution allocation module is configured to determine, based on multiple performance characteristics of each participant in the current match, the degree of contribution of the performance characteristics to the match prediction result output by the match prediction model, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the statistics module is configured to summarize the contribution levels corresponding to the multiple performance characteristics respectively, and obtain the contribution evaluation result of each participant in this game;
  • the evaluation module is configured to determine the target of this competition from the participants based on the contribution evaluation results.
  • embodiments of this disclosure provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for determining the target object.
  • embodiments of this disclosure provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the above-described method for determining the target object.
  • Figure 1 shows a flowchart illustrating a method for determining a target object provided in one embodiment of this disclosure
  • Figure 3 shows a flowchart illustrating a method for optimizing the selection of target objects based on multiple competition characteristics provided in one embodiment of this disclosure
  • Figure 4 shows a schematic diagram of the structure of a target object determination device provided in one embodiment of the present disclosure
  • Figure 5 is a schematic diagram of the structure of an electronic device 500 provided in one embodiment of the present disclosure.
  • the present disclosure provides a method, apparatus, device and storage medium for determining target objects.
  • it provides a more fair, objective and reasonable basis for the selection of target objects, making the target object selection results simple, transparent and highly interpretable. This is beneficial to help various event organizers to more comprehensively and objectively evaluate the performance of each participant.
  • the method includes steps S101-S103; specifically:
  • the method for determining the target object determines the contribution of each performance feature to the prediction result output by the prediction model, given that the model input data includes the performance features, based on multiple performance features of each participant in the match.
  • the contribution levels corresponding to each performance feature are then summarized to obtain a contribution evaluation result for each participant in the match.
  • the target object for the match is determined from among the participants.
  • this match refers to the match in which the target object needs to be selected; this match can be an online match (e.g., a game session in a game) or an offline match (e.g., a basketball game held offline).
  • the target object can refer to the best performing participant in this match (e.g., the MVP of this match) or the participant who makes the highest contribution to their team in this match. This disclosed embodiment does not limit the specific match type of this match or the specific meaning of the target object to be determined in this match.
  • the performance data of that participant in the match can be obtained from the match data recorded in the match. Then, based on the various match characteristics that need to be referenced when determining the target, the target match performance data that matches each match characteristic can be extracted from the participant's match performance data as multiple performance characteristics of that participant in the match; that is, one performance characteristic can correspond to one match characteristic, and the specific characteristic value of a performance characteristic can be determined based on the target match performance data that matches the above match characteristics.
  • the participants can represent each player participating in the game.
  • the aforementioned performance data can represent the player's performance data in this game.
  • various game characteristics that need to be considered may include: the cumulative number of times the player kills the enemy game character, the cumulative game damage dealt to the enemy game team, etc.
  • the multiple performance characteristics of each player in this game may include: the cumulative number of times the player kills the enemy game character, the cumulative game damage dealt to the enemy game team, etc.
  • this game when this game is an online or offline basketball game, the participants can represent each basketball player participating in the game.
  • the above-mentioned performance data can represent the performance data of the basketball players in this game.
  • various game characteristics that need to be considered may include: cumulative points, cumulative assists, cumulative rebounds, cumulative steals, cumulative blocks, etc.
  • the multiple performance characteristics of each basketball player in this game may include: each basketball player's cumulative points, cumulative assists, cumulative rebounds, cumulative steals, cumulative blocks, etc.
  • the match result prediction model represents a classification model pre-trained based on historical match data. For example, if there are two participating teams in each match (referred to as the first participating team and the second participating team), the performance data of each participating team in a match (i.e., the match data of a match) is input into the match result prediction model.
  • the match result prediction model predicts the match result of the match.
  • the predicted match result can be: the probability of the first participating team winning (which is also equivalent to the probability of the second participating team losing) and the probability of the first participating team losing (which is also equivalent to the probability of the second participating team winning).
  • the match result prediction model when predicting the match result of this match using the match result prediction model, all performance characteristics of all participants in this match (i.e., the performance data of all participants in this match) should originally be used as the model input data of the match result prediction model. That is, the match prediction result output by the match result prediction model can be interpreted as the result of the combined effect of each performance characteristic of each participant in this match. Based on this, when executing step S101, for each participant's performance characteristic in this match, the difference between the match prediction results obtained when the model input data includes the performance characteristic and when it does not can be used to determine the degree of contribution of the performance characteristic to the above match prediction result.
  • the calculation method of the Shapley value can be combined with the match result prediction process of the match result prediction model in the manner described in step a.
  • the contribution of each performance feature to the match prediction result can be quantitatively represented. Specifically:
  • Step a Use the score function in the match result prediction model as the utility function in the Shapley value calculation formula, and calculate the Shapley value of the performance feature as the degree of contribution of the performance feature to the match prediction result output by the match result prediction model.
  • the above-mentioned scoring function represents the function used in the competition result prediction model to output the competition prediction result based on the model input data; wherein, the scoring function used in the competition result prediction model may also be different depending on the model structure to which the competition result prediction model belongs, and this disclosure embodiment does not limit the specific function type to which the above-mentioned scoring function belongs.
  • step a taking the original model input data of the competition result prediction model as having n performance features (equivalent to the sum of n performance features of all participants) as an example, the Shapley value of the j-th performance feature (that is, the contribution of the j-th performance feature to the competition prediction result output by the competition result prediction model) can be calculated according to the following improved Shapley value calculation formula:
  • N represents the set of performance features consisting of n performance features
  • S represents any subset that can be formed by all other performance features in the aforementioned performance feature set N, except for the j-th performance feature;
  • v represents the score function in the match result prediction model (i.e., the utility function in the Shapley value calculation formula);
  • v(S ⁇ j ⁇ ) represents the match prediction result output by the match prediction model when the model input data is a subset S and the j-th performance feature (specifically, it can be the probability of the target team winning in this match, where the target team can represent any team in this match).
  • v(S) represents the prediction result of the match result prediction model when the model input data is a subset S (or the probability of the target participating team winning).
  • ⁇ j (v) represents the Shapley value of the j-th performance feature (that is, the degree of contribution of the j-th performance feature to the match prediction result output by the match result prediction model).
  • the number of performance features contained in S ⁇ j ⁇ and subset S may be less than the performance feature set N when calculating v(S ⁇ j ⁇ ) and v(S).
  • the number of performance features contained in S ⁇ j ⁇ and subset S can be calculated based on S ⁇ j ⁇ and subset S.
  • v(S ⁇ j ⁇ ) and v(S) zeros are padded to represent the multiple features missing from the set S relative to the set N.
  • the training data e.g., historical match data
  • the feature values of the missing features in the training data can be substituted into the calculation process of v(S ⁇ j ⁇ ) and v(S).
  • the mean of v(S ⁇ j ⁇ ) and the mean of v(S) calculated multiple times can be used as the final v(S ⁇ j ⁇ ) and v(S).
  • the shapley value corresponding to each performance feature (i.e., the degree of contribution of each performance feature to the match prediction result output by the match result prediction model) can be calculated separately. Since each performance feature corresponds to a specific participant, the shapley values corresponding to all performance features of the same participant can be summarized to obtain the contribution evaluation result of a participant in this match.
  • the match data can be used as follows: In the match data, the performance characteristics of multiple participants belonging to the first participating team are ranked first, and the performance characteristics of multiple participants belonging to the second participating team are ranked last. The ranked match data is then input into the match result prediction model. Based on the match prediction result output by the model, which is the probability of the first participating team winning, the sum of the Shapley values (denoted as M1) of multiple performance characteristics of a participant (denoted as participant m) is calculated.
  • the performance characteristics of multiple participants belonging to the first participating team are ranked last, and the performance characteristics of multiple participants belonging to the second participating team are ranked last.
  • the performance characteristics of multiple participants in a team are ranked first.
  • the ranked match data is input into the match result prediction model.
  • the sum of the Shapley values (denoted as M2) of multiple performance characteristics of a participant (denoted as participant m) is calculated.
  • the contribution evaluation result of participant m in this match can be summarized as M1-M2 (equivalent to the higher the contribution of participant m's performance characteristics to the match prediction result when the match prediction result is the probability of the opponent's second participating team winning, the lower its contribution to its own first participating team).
  • the contribution evaluation result of participant m in this match can be summarized as M2-M1.
  • the model when the model outputs a match prediction result representing the probability of the participating object's own team winning, the higher the contribution of the participating object's performance characteristics to the match prediction result, the higher its contribution to its own team; when the model outputs a match prediction result representing the probability of the participating object's opponent's team winning, the higher the contribution of the participating object's performance characteristics to the match prediction result, the lower its contribution to its own team (i.e., when summarizing, the Shapley values of the participating object's multiple performance characteristics should be their inverses).
  • the larger the value of the contribution evaluation result the greater the contribution of the participant's performance in this game (i.e., the performance characteristics mentioned above) to the game prediction result output by the game result prediction model.
  • the larger the value of the contribution evaluation result the more crucial the role of the participant in the winning team's victory (i.e., the game result) in this game. Therefore, the participant with the largest value of the above contribution evaluation result (or the highest ranking of the above contribution evaluation result) can be identified as the target object of this game (such as the MVP of this game).
  • the participants in this competition can be sorted in descending order of the above contribution evaluation results, and the participant with the highest ranking in the above contribution evaluation results can be determined as the target of this competition.
  • the participants in the winning team can be sorted in descending order of the above contribution evaluation results, and the participant with the highest contribution evaluation result in the winning team can be determined as the target of this competition.
  • Figure 2 shows a flowchart of a method for training a match result prediction model provided by an embodiment of this disclosure. As shown in Figure 2, before executing step S101, the method includes the following steps S201-S202, specifically:
  • the match data for each match includes multiple historical performance characteristics of multiple historical participants in that match; for a detailed explanation of the match data and performance characteristics, please refer to the relevant explanation in step S101 above, and the repetitions will not be repeated here.
  • the competition result prediction model can be the LightGBM model or a classification model with other model structures. This disclosure does not limit the specific model structure to which the competition result prediction model belongs.
  • a tree-structured match result prediction model (such as the LightGBM model mentioned above) can be preferentially selected.
  • the match result prediction model can be trained based on the classification loss between the match prediction model output and the actual match result, until the match result prediction model converges, resulting in a match result prediction model including the adjusted model parameters.
  • step S101 when executing step S101, based on the multiple competition features that need to be considered when selecting target objects, target competition performance data matching each competition feature can be extracted from the competition performance data of the participant as multiple performance features of the participant in this competition; wherein, in order to optimize the multiple competition features that need to be considered when selecting target objects, Figure 3 shows a flowchart of a method for optimizing multiple competition features that need to be considered when selecting target objects provided by an embodiment of this disclosure. As shown in Figure 3, before executing step S101, the method includes the following steps S301-S304, specifically:
  • the scoring function in the match result prediction model is used as the utility function in the shapley value calculation formula to calculate the shapley values corresponding to various match features in the same historical match.
  • the specific calculation method of the Shapley value corresponding to each competition feature in step S301 can refer to the specific calculation method of the Shapley value corresponding to each performance feature in step a above. The repetition will not be repeated here.
  • the average shapley value of match feature y1 in the m historical matches can be calculated based on the m shapley values corresponding to match feature y1 in the m historical matches (i.e., the average of the above m shapley values).
  • the competition feature that meets the target screening criteria can be: the competition feature whose average Shapley value is greater than or equal to a preset threshold.
  • the competition feature that meets the target screening criteria can also be: the average Shapley value is greater than any other competition feature and the difference between the average Shapley value and any other competition feature is greater than or equal to a preset difference threshold (equivalent to a competition feature whose average Shapley value is much greater than other competition features among the above multiple competition features).
  • the competition features that need to be deleted are competition features with significantly larger average Shapley values.
  • the calculated average Shapley value can characterize the importance value of a competition feature among the above multiple competition features.
  • the competition feature may lead to label leakage, which reduces the overall calculation accuracy of the Shapley values corresponding to multiple competition features. Therefore, the competition feature can be deleted in order to maintain the overall calculation accuracy of the Shapley values corresponding to multiple competition features.
  • step S304 can be referred to the relevant description of competition characteristics and performance characteristics in the aforementioned step S101, and the repeated parts will not be repeated here.
  • the target game feature includes the cumulative number of rebounds
  • the cumulative number of rebounds obtained by each participant in the game can be determined from the performance data of each participant in the game as a performance feature of each participant in the game.
  • this disclosure also provides the following three optional implementation methods for determining the target object of the season (such as determining the MVP of the season) at the end of the season:
  • the target group for this season can be determined in the manner described in steps b1-b2 below, specifically:
  • Step b1 At the end of the season, obtain the ranking of the average contribution evaluation results of the participants in each of the winning games of the season as the first average contribution evaluation result ranking.
  • step b1 the target object determination method shown in steps S101-S103 above can be used to obtain the contribution evaluation result ranking of each participant in each game of this season. Then, for each participant, multiple contribution evaluation result rankings of the participant in the winning games can be filtered out from the obtained contribution evaluation result rankings in each game, and the average of the above multiple contribution evaluation result rankings can be calculated as the first average contribution evaluation result ranking of the participant.
  • Step b2 From the participants in this season, determine the participant with the highest ranking in the first average contribution evaluation as the target participant for this season.
  • the ranking of the first average contribution evaluation result represents the ranking of a participant's average contribution evaluation result in each of the winning games this season. That is, the higher the ranking of the first average contribution evaluation result (i.e. the smaller the value), the greater the contribution of the participant to the results of each winning game, and the more qualified they are to become the target of this season (such as this season's MVP).
  • the target group for this season can be determined in the manner described in steps c1-c2 below, specifically:
  • Step c1 At the end of the season, obtain the ranking of the average contribution evaluation results of the participants in each game of the season as the second average contribution evaluation result ranking.
  • step c1 can refer to the specific implementation of step b1 above, except that the number of winning matches in step b1 is replaced with all the matches participated in. The repetitions will not be repeated here.
  • Step c2 From the participants of this season, determine the participant with the highest ranking in the second average contribution evaluation as the target participant for this season.
  • the ranking of the second average contribution evaluation results represents the ranking of a participant's average contribution evaluation results in each game of the season. That is, the higher the ranking of the second average contribution evaluation results (i.e. the smaller the value), the greater the contribution of the participant to the game results in each game, and the more qualified they are to become the target of this season (such as this season's MVP).
  • the target objects for this season can be determined in the manner described in steps d1-d2 below, specifically:
  • Step d1 At the end of the season, obtain the average contribution evaluation results of the participants in each game of the season.
  • step d1 the target object determination method shown in steps S101-S103 above can be used to obtain the contribution evaluation results of each participant in each game of this season. Then, for each participant, the average value of the above contribution evaluation results is calculated based on the number of games the participant has participated in, and is used as the average contribution evaluation result of the participant in each game of this season.
  • Step d2 From the participants in this season, determine the participant with the highest average contribution evaluation result as the target participant for this season.
  • the average contribution evaluation result represents the average contribution evaluation result of a participant in each game of the season. That is, the higher the value of the average contribution evaluation result, the greater the contribution of the participant to the game results in each game, and the more qualified they are to become the target of the season (such as the MVP of the season).
  • the contribution of each performance characteristic to the match prediction result output by the match prediction model is determined, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the contribution levels corresponding to multiple performance characteristics are summarized to obtain the contribution evaluation result of each participant in the match.
  • the target object for the match is determined from the participants. In this way, this disclosure provides a more fair, objective, and reasonable basis for the selection of target objects, making the target object selection results simple, transparent, and highly interpretable, which is conducive to helping various event organizers to more comprehensively and objectively evaluate the performance of each participant in the match.
  • this disclosure also provides an apparatus corresponding to the above-described method for determining the target object. Since the principle of the apparatus for determining the target object in the embodiments of this disclosure is similar to that of the above-described method for determining the target object in the embodiments of this disclosure, the implementation of the apparatus for determining the target object can refer to the implementation of the above-described method for determining the target object, and the repeated parts will not be described again.
  • FIG4 shows a schematic diagram of the structure of a target object determining device provided in an embodiment of the present disclosure, wherein the determining device includes:
  • the contribution allocation module 401 is configured to determine, based on multiple performance characteristics of each participant in the current match, the degree of contribution of the performance characteristics to the match prediction result output by the match prediction model, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the statistics module 402 is configured to summarize the contribution levels corresponding to the multiple performance characteristics respectively, and obtain the contribution evaluation result of each participant in this game.
  • the evaluation module 403 is configured to determine the target of this competition from the participants based on the contribution evaluation results.
  • the contribution allocation module 401 when determining the contribution of each performance feature to the match prediction result output by the match prediction model, based on multiple performance features of each participant in the match, and assuming that the performance features are included in the model input data of the match result prediction model, the contribution allocation module 401 is configured as follows:
  • the scoring function in the match result prediction model is used as the utility function in the Shapley value calculation formula.
  • the Shapley value of the performance feature is calculated as the degree of contribution of the performance feature to the match prediction result output by the match result prediction model.
  • the scoring function represents the function in the match result prediction model used to output the match prediction result based on the model input data.
  • the apparatus further includes a feature filtering module, wherein the feature filtering module is configured to:
  • the apparatus further includes a model training module, wherein the model training module is configured to pre-train the match result prediction model using the following method:
  • the match data from multiple matches is input into the match result prediction model, and the match result prediction model outputs the match prediction results for each match; wherein, the match data includes multiple historical performance characteristics of multiple historical participants in the current match;
  • the match result prediction model is trained based on the loss between the predicted result and the actual result of the same match until the match result prediction model converges.
  • the participants in this competition are ranked according to their contribution evaluation results from highest to lowest, and the participant with the highest contribution evaluation result is determined as the target of this competition.
  • the participant with the highest ranking in the first average contribution assessment will be selected as the target for this season.
  • the season settlement module is further configured to:
  • the ranking of the average contribution evaluation results of the participants in each game of the season will be used as the second average contribution evaluation result ranking.
  • the participant with the highest ranking in the second average contribution assessment will be selected as the target for this season.
  • the season settlement module is further configured to:
  • the participant with the highest average contribution evaluation result is selected as the target for this season.
  • the contribution of each participant to the match prediction model's output match prediction result is determined according to multiple performance characteristics of each participant in the match, provided that the model input data of the match result prediction model includes performance characteristics.
  • the contribution levels corresponding to each of the multiple performance characteristics are summarized to obtain the contribution evaluation result of each participant in the match.
  • the target object for the match is determined from the participants.
  • this disclosure also provides an electronic device corresponding to the above-mentioned method for determining the target object. Since the principle of solving the problem by the electronic device in the embodiments of this disclosure is similar to the above-mentioned method for determining the target object in the embodiments of this disclosure, the implementation of the electronic device can refer to the implementation of the above-mentioned method for determining the target object, and the repeated parts will not be described again.
  • Figure 5 is a schematic diagram of the structure of an electronic device 500 provided in an embodiment of this disclosure, including: a processor 501, a memory 502, and a bus 503.
  • the memory 502 stores machine-readable instructions executable by the processor 501.
  • the processor 501 communicates with the memory 502 via the bus 503.
  • the processor 501 executes the machine-readable instructions, and when the processor 501 executes the machine-readable instructions, it implements the following steps:
  • the contribution of the performance characteristics to the match prediction results output by the match prediction model is determined, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the target of this competition is determined from the participants.
  • the processor 501 when determining the contribution of each performance feature to the match prediction result output by the match prediction model, based on multiple performance features of each participant in the match, and assuming that the performance features are included in the model input data of the match result prediction model, the processor 501 is configured to:
  • the scoring function in the match result prediction model is used as the utility function in the Shapley value calculation formula.
  • the Shapley value of the performance feature is calculated as the degree of contribution of the performance feature to the match prediction result output by the match result prediction model.
  • the scoring function represents the function in the match result prediction model used to output the match prediction result based on the model input data.
  • the processor 501 before determining the contribution of each performance feature to the match prediction result output by the match prediction model, based on multiple performance features of each participant in the match, and assuming that the performance features are included in the model input data of the match result prediction model, the processor 501 is configured to:
  • the scoring function in the match result prediction model is used as the utility function in the shapley value calculation formula to calculate the shapley values corresponding to various match characteristics in the same historical match.
  • the competition features whose average Shapley value meets the target screening criteria are removed from the multiple competition features to obtain the remaining target competition features;
  • processor 501 is configured to pre-train the match result prediction model using the following method:
  • the match data from multiple matches is input into the match result prediction model, and the match result prediction model outputs the match prediction results for each match; wherein, the match data includes multiple historical performance characteristics of multiple historical participants in the current match;
  • the match result prediction model is trained based on the loss between the predicted result and the actual result of the same match until the match result prediction model converges.
  • the processor 501 when determining the target of the competition from the participants based on the contribution evaluation results, is configured to:
  • the participants in this competition are ranked according to their contribution evaluation results from highest to lowest, and the participant with the highest contribution evaluation result is determined as the target of this competition.
  • processor 501 is further configured to:
  • the ranking of the average contribution evaluation results of the participants in each of the winning games of the season will be used as the first average contribution evaluation result ranking.
  • the participant with the highest ranking in the first average contribution assessment will be selected as the target for this season.
  • processor 501 is further configured to:
  • the ranking of the average contribution evaluation results of the participants in each game of the season will be used as the second average contribution evaluation result ranking.
  • the participant with the highest ranking in the second average contribution assessment will be selected as the target for this season.
  • processor 501 is further configured to:
  • the participant with the highest average contribution evaluation result is selected as the target for this season.
  • the electronic device provided in this disclosure determines the contribution of these performance characteristics to the match prediction model's output, assuming the model input data includes these characteristics.
  • the contribution levels of each performance characteristic are then summarized to obtain a contribution evaluation result for each participant in the match.
  • the target participant for the match is determined from among the participants.
  • this disclosure also provides a computer-readable storage medium storing a computer program, which is executed by a processor, wherein the processor performs the following steps:
  • the contribution of the performance characteristics to the match prediction results output by the match prediction model is determined, provided that the model input data of the match result prediction model includes the performance characteristics.
  • the target of this competition is determined from the participants.
  • the processor when determining the contribution of each performance feature to the match prediction model's output match prediction result, based on multiple performance features of each participant in the match, and assuming these performance features are included in the model input data of the match result prediction model, the processor is configured to:
  • the scoring function in the match result prediction model is used as the utility function in the Shapley value calculation formula.
  • the Shapley value of the performance feature is calculated as the degree of contribution of the performance feature to the match prediction result output by the match result prediction model.
  • the scoring function represents the function in the match result prediction model used to output the match prediction result based on the model input data.
  • the processor before determining the contribution of the performance features to the match prediction model output by the match prediction model, based on multiple performance features of each participant in the match, provided that the performance features are included in the model input data of the match result prediction model, the processor is configured to:
  • the scoring function in the match result prediction model is used as the utility function in the shapley value calculation formula to calculate the shapley values corresponding to various match characteristics in the same historical match.
  • the competition features whose average Shapley value meets the target screening criteria are removed from the multiple competition features to obtain the remaining target competition features;
  • the processor is configured to pre-train the match result prediction model using the following method:
  • the match data from multiple matches is input into the match result prediction model, and the match result prediction model outputs the match prediction results for each match; wherein, the match data includes multiple historical performance characteristics of multiple historical participants in the current match;
  • the match result prediction model is trained based on the loss between the predicted result and the actual result of the same match until the match result prediction model converges.
  • the processor when determining the target for the match from the participants based on the contribution evaluation results, the processor is configured to:
  • the participants in this competition are ranked according to their contribution evaluation results from highest to lowest, and the participant with the highest contribution evaluation result is determined as the target of this competition.
  • the processor is further configured to:
  • the ranking of the average contribution evaluation results of the participants in each of the winning games of the season will be used as the first average contribution evaluation result ranking.
  • the participant with the highest ranking in the first average contribution assessment will be selected as the target for this season.
  • the processor is further configured to:
  • the ranking of the average contribution evaluation results of the participants in each game of the season will be used as the second average contribution evaluation result ranking.
  • the participant with the highest ranking in the second average contribution assessment will be selected as the target for this season.
  • the processor is further configured to:
  • the participant with the highest average contribution evaluation result is selected as the target for this season.
  • the contribution of each performance characteristic to the match prediction model's output match prediction result is determined, provided that the model input data of the match prediction model includes the performance characteristics.
  • the contribution levels corresponding to each performance characteristic are summarized to obtain the contribution evaluation result of each participant in the match.
  • the target object for the match is determined from among the participants.
  • the computer-readable storage medium can also execute other machine-readable instructions when the processor runs, to perform the target object determination method as described in other embodiments.
  • the specific steps and principles of the target object determination method please refer to the description of the method-side embodiment, which will not be repeated here.
  • the units described as separate components may or may not be physically separate.
  • the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
  • the functional units in the embodiments provided in this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
  • the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
  • This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure.
  • the aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Computer Security & Cryptography (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种目标对象的确定方法、装置、设备及存储介质,该确定方法包括:根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度(S101);对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果(S102);根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象(S103)。为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。

Description

一种目标对象的确定方法、装置、设备及存储介质
相关申请交叉引用
本申请要求于2024年7月19日提交的申请号为:202410979559.7、名称为“一种目标对象的确定方法、装置、设备及存储介质”的中国专利申请的优先权,该中国专利申请的全部内容通过引用全部并入本文。
技术领域
本公开涉及数据处理技术领域,具体而言,涉及一种目标对象的确定方法、装置、设备及存储介质。
背景技术
在比赛结束之后,常常会需要根据每个参赛对象的赛场表现,从所有的参赛对象中评选出本场比赛表现最佳的目标对象,例如,目标对象可以是本场比赛的MVP(Most Valuable Player,最有价值选手)选手。
现有的目标对象评选方案可以分为两类:第一类是根据参赛对象在比赛中的多种表现特征以及每种表现特征对应的权重系数,通过加权求和的方式得到各参赛对象的赛场表现得分,从中选择赛场表现得分最高的参赛对象作为当场比赛表现最佳的目标对象(多用于线上比赛使用);第二类是通过专家投票的方式,确定获得专家票数最高的参赛对象成为当场比赛表现最佳的目标对象(多用于线下比赛使用)。
但是上述两类目标对象评选方案,由于不同表现特征对应的权重系数大小以及专家的具体投票本质上都依赖于人为经验,从而导致最终得出的目标对象评选结果无法使用更加客观的数据进行解释,使得目标对象评选结果存在受到人为因素干扰较大且可解释性较差的缺陷。
发明内容
有鉴于此,本公开提供一种目标对象的确定方法、装置、设备及存储介质,通过改进现有的目标对象评选方式,为目标对象的评选提供更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
为使本公开的上述目的、特征和优点能更明显易懂,下文特举较佳实施例,并配合所附附图,作详细说明如下。
第一方面,本公开实施例提供了一种目标对象的确定方法,所述确定方法包括:
根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
第二方面,本公开实施例提供了一种目标对象的确定装置,所述确定装置包括:
贡献分配模块,被配置为根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
统计模块,被配置为对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
评选模块,被配置为根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
第三方面,本公开实施例提供了一种计算机设备,包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时实现上述的目标对象的确定方法的步骤。
第四方面,本公开实施例提供了一种计算机可读存储介质,所述计算机可读存储介质上存储有计算机程序,所述计算机程序被处理器运行时执行上述的目标对象的确定方法的步骤。
本公开的实施例提供的技术方案可以包括以下有益效果:
本公开实施例提供的一种目标对象的确定方法、装置、设备及存储介质,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
附图说明
为了更清楚地说明本公开实施例的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,应当理解,以下附图仅示出了本公开的某些实施例,因此不应被看作是对范围的限定,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他相关的附图。
图1示出了本公开其中之一实施例所提供的一种目标对象的确定方法的流程示意图;
图2示出了本公开其中之一实施例提供的一种训练得到比赛结果预测模型的方法的流程示意图;
图3示出了本公开其中之一实施例提供的一种优化评选目标对象时需要参考的多种比赛特征的方法的流程示意图;
图4示出了本公开其中之一实施例所提供的一种目标对象的确定装置的结构示意图;
图5为本公开其中之一实施例提供的一种电子设备500的结构示意图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚,下面将结合本公开实施例中的附图,对本公开实施例中的技术方案进行清楚、完整地描述,应当理解,本公开中附图仅起到说明和描述的目的,并不用于限定本公开的保护范围。另外,应当理解,示意性的附图并未按实物比例绘制。本公开中使用的流程图示出了根据本公开的一些实施例实现的操作。应该理解,流程图的操作可以不按顺序实现,没有逻辑的上下文关系的步骤可以反转顺序或者同时实施。此外,本领域技术人员在本公开内容的指引下,可以向流程图添加一个或多个其他操作,也可以从流程图中移除一个或多个操作。
另外,所描述的实施例仅仅是本公开一部分实施例,而不是全部的实施例。通常在此处附图中描述和示出的本公开实施例的组件可以以各种不同的配置来布置和设计。因此,以下对在附图中提供的本公开的实施例的详细描述并非旨在限制要求保护的本公开的范围,而是仅仅表示本公开的选定实施例。基于本公开的实施例,本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例,都属于本公开保护的范围。
需要说明的是,本公开实施例中将会用到术语“包括”,用于指出其后所声明的特征的存在,但并不排除增加其它的特征。
现有的两类目标对象评选方案,由于不同表现特征对应的权重系数大小以及专家的具体投票本质上都依赖于人为经验,从而导致最终得出的目标对象评选结果无法使用更加客观的数据进行解释,使得目标对象评选结果存在受到人为因素干扰较大且可解释性较差的缺陷。
基于此,本公开实施例提供了一种目标对象的确定方法、装置、设备及存储介质,通过改进现有的目标对象评选方式,为目标对象的评选提供更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
在本公开其中一种实施例中的一种目标对象的确定方法可以运行于终端设备或者是服务器。其中,终端设备可以为本地终端设备。当目标对象的确定方法运行于服务器时,该确定方法则可以基于云交互系统来实现与执行,其中,云交互系统包括服务器和客户端设备(也即终端设备)。
为便于对本公开实施例进行理解,下面对本公开实施例提供的一种目标对象的确定方法、装置、设备及存储介质进行详细介绍。
参照图1所示,图1示出了本公开实施例所提供的一种目标对象的确定方法的流程示意图,其中,所述确定方法包括步骤S101-S103;具体的:
S101,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度。
S102,对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果。
S103,根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
本公开实施例提供的上述目标对象的确定方法,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
下面对本公开实施例提供的上述目标对象的确定方法中的各步骤分别进行示例性的说明:
S101,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度。
这里,本场比赛表征当前需要评选目标对象的比赛;其中,本场比赛可以属于线上比赛(如,游戏中的一场游戏对局),也可以属于线下比赛(如,线下进行的一场篮球赛),上述目标对象可以表征本场比赛中表现最佳的参赛对象(如可以是本场比赛的MVP),也可以表征本场比赛中对于自身所在队伍作出最高贡献的参赛对象,对于本场比赛所属的具体比赛类型以及本场比赛中需要确定的上述目标对象所表征的具体含义,本公开实施例不作任何限定。
这里,在本场比赛结束后,针对每个参赛对象,可以从本场比赛记录的比赛数据中,获取该名参赛对象在本场比赛中的赛场表现数据,然后,根据确定目标对象时需要参考的多种比赛特征,可以从该名参赛对象的赛场表现数据中,提取与每种比赛特征相匹配的目标赛场表现数据作为该名参赛对象在本场比赛中的多个表现特征;也即一个表现特征可以对应一种比赛特征,而一个表现特征的具体特征值则可以根据与上述比赛特征相匹配的目标赛场表现数据确定。
需要说明的是,在不同类型的比赛中,上述表现特征表示的具体特征含义可能会有所差异,对于上述表现特征所表征的具体特征含义以及每名参赛对象在本场比赛中的具体表现特征数量,本公开实施例不作任何限定。
示例性的说明,当本场比赛属于线上的一场游戏对局时,则参赛对象可以表征参与本场游戏对局的每个玩家,上述赛场表现数据可以表征玩家在本场游戏对局中的游戏表现数据,在游戏对局中评选MVP(即评选目标对象)时需要参考的多种比赛特征可以包括:击杀敌方游戏角色的累计次数、对敌方游戏队伍造成的累计游戏伤害等,每名玩家在本场游戏对局中的多个表现特征可以包括:每名玩家在本场游戏对局击杀敌方游戏角色的累计次数、对敌方游戏队伍造成的累计游戏伤害等。
示例性的说明,当本场比赛属于线上或者线下的一场篮球赛时,则参赛对象可以表征参与本场比赛的每个篮球选手,上述赛场表现数据可以表征篮球选手在本场比赛中的赛场表现数据,在本场比赛中评选MVP(即评选目标对象)时需要参考的多种比赛特征可以包括:累计得分数、累计助攻次数、累计篮板球数量、累计抢断次数、累计获得的盖帽数量等,每个篮球选手在本场比赛中的多个表现特征可以包括:每个篮球选手在本场比赛中的累计得分数、累计助攻次数、累计篮板球数量、累计抢断次数、累计获得的盖帽数量等。
这里,在步骤S101中,比赛结果预测模型表征根据历史场次的比赛数据预先训练好的分类模型;其中,以每场比赛的参赛队伍数量是2队(记作第一参赛队伍和第二参赛队伍)为例,将一场比赛中各参赛对象的赛场表现数据(即一场比赛的比赛数据)输入比赛结果预测模型,通过比赛结果预测模型对该场比赛的比赛结果进行预测,得到的比赛预测结果可以是:第一参赛队伍胜利的概率(也相当于第二参赛队伍失败的概率)和第一参赛队伍失败的概率(也相当于第二参赛队伍胜利的概率)。
在本公开实施例中,参考上述关于比赛结果预测模型的相关说明内容可知,在通过比赛结果预测模型对本场比赛的比赛结果进行预测时,原本应该将所有参赛对象在本场比赛中的全部表现特征(即所有参赛对象在本场比赛中的赛场表现数据)作为比赛结果预测模型的模型输入数据,也即,比赛结果预测模型输出的比赛预测结果可以解释成:每个参赛对象在本场比赛中的每个表现特征共同作用的结果,基于此,在执行步骤S101时,可以针对每个参赛对象在本场比赛中的一个表现特征,通过模型输入数据中包含该表现特征与不包含该表现特征时得到的比赛预测结果差异,来确定当模型输入数据中包含该表现特征时,该表现特征能够对上述比赛预测结果产生的贡献程度。
这里,考虑到shapley值是唯一满足四种理想性质(有效性、对称性、可加性和无效性)的公平贡献分配方式,作为一可选实施例,可以按照以下步骤a所述的方式,将shapley值的计算方式与比赛结果预测模型的比赛结果预测过程相结合,通过计算出的每个参赛对象在本场比赛中的每个表现特征对应的shapley值,来量化地表示上述每个表现特征对比赛预测结果的贡献程度,具体的:
步骤a、将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算所述表现特征的shapley值作为所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度。
这里,上述得分函数表征比赛结果预测模型中用于根据模型输入数据输出比赛预测结果的函数;其中,比赛结果预测模型所属的模型结构不同,比赛结果预测模型中使用的得分函数也可能存在差异,对于上述得分函数所属的具体函数类型,本公开实施例不作任何限定。
具体的,在步骤a中,以比赛结果预测模型原本对应的模型输入数据是n个表现特征(相当于所有参赛对象的表现特征之和是n个)为例,可以按照如下改进后的shapley值计算式,计算得到第j个表现特征的shapley值(也即第j个表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度):
其中,N表示n个表现特征组成的表现特征集合;
S表示上述表现特征集合N中除第j个表现特征之外的所有其他表现特征可以组成的任意一个子集;
|S|表示上述子集S中的表现特征数量;
v表示比赛结果预测模型中的得分函数(也即shapley值计算式中的效用函数);
v(S∪{j})表示比赛结果预测模型的模型输入数据是子集S和第j个表现特征时,比赛结果预测模型输出的比赛预测结果(具体可以是本场比赛中目标参赛队伍胜利的概率,其中,目标参赛队伍可以表征本场比赛中的任意一个参赛队伍);
v(S)表示比赛结果预测模型的模型输入数据是子集S时,比赛结果预测模型输出的比赛预测结果(也可以是上述目标参赛队伍胜利的概率);
φj(v)表示第j个表现特征的shapley值(也即第j个表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度)。
需要说明的是,针对上述shapley值计算式中的v(S∪{j})和v(S),由于比赛结果预测模型中在进行比赛结果预测时原本使用的模型输入数据是表现特征集合N,因此,在计算上述v(S∪{j})和v(S)时,基于S∪{j}和子集S中包含的表现特征数量可能会小于表现特征集合N,为弥补因模型输入数据中表现特征数量的减少而可能会对模型输出的比赛预测结果产生的影响,作为一可选实施例,可以根据S∪{j}和子集S中相对于表现特征集合N所缺少的多个表现特征,在计算v(S∪{j})和v(S)时,对所缺少的多个表征特征进行补零;作为另一可选实施例,则还可以参考比赛结果预测模型在模型训练阶段使用的训练数据(如,历史比赛数据),将所缺少的特征在上述训练数据中的特征值代入到v(S∪{j})和v(S)的计算过程中,并使用多次计算出的v(S∪{j})的均值和v(S)的均值作为最终得到的v(S∪{j})和v(S)。
S102,对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果。
这里,参考前述步骤S101中步骤a所示的shapley值计算方式,可以分别计算出每个表现特征对应的shapley值(即每个表现特征对比赛结果预测模型输出的比赛预测结果的贡献程度),基于每个表现特征对应有具体的参赛对象,因此可以对同一个参赛对象的所有表现特征对应的shapley值进行汇总,即可得到一个参赛对象在本场比赛中的贡献度评估结果。
需要说明的是,参考上述shapley值计算式中关于比赛预测结果的相关说明内容,在对一个参赛对象的多个表现特征对应的shapley值进行汇总时,若v(S∪{j})和v(S)表示的比赛预测结果是该参赛对象所在参赛队伍获胜的概率,则该参赛对象的表现特征对于比赛预测结果的贡献程度越高表示其对自身所在参赛队伍作出的贡献越高,可以直接对该参赛对象的所有表现特征对应的shapley值进行求和,得到该参赛对象的在本场比赛中的贡献度评估结果;若v(S∪{j})和v(S)表示的比赛预测结果是该参赛对象对手所在参赛队伍获胜的概率,则该参赛对象的表现特征对于比赛预测结果的贡献程度越高反而表示其对自身所在参赛队伍作出的贡献越低,此时需要对该参赛对象的所有表现特征对应的shapley值的相反数进行求和,得到该参赛对象的在本场比赛中的贡献度评估结果。
除此之外,作为另一可选实施例,为进一步提高同一个参赛对象的上述贡献度评估结果的准确度,以本场比赛共有两个参赛队伍(记作第一参赛队伍和第二参赛队伍)为例,还可以针对本场比赛的比赛数据,在上述比赛数据中,将属于第一参赛队伍的多个参赛对象的表现特征排序在前,将属于第二参赛队伍的多个参赛对象的表现特征排序在后,将排序后的比赛数据输入比赛结果预测模型中,按照模型输出的比赛预测结果是上述第一参赛队伍胜利的概率,计算出一个参赛对象(记作参赛对象m)的多个表现特征的shapley值的和值(记作M1);然后,在上述比赛数据中,将属于第一参赛队伍的多个参赛对象的表现特征排序在后,将属于第二参赛队伍的多个参赛对象的表现特征排序在前,将排序后的比赛数据输入比赛结果预测模型中,按照模型输出的比赛预测结果是上述第二参赛队伍胜利的概率,计算出一个参赛对象(记作参赛对象m)的多个表现特征的shapley值的和值(记作M2);其中,当参赛对象m属于第一参赛队伍时,则可以汇总得到参赛对象m在本场比赛中的贡献度评估结果是M1-M2(相当于在比赛预测结果是对手所在第二参赛队伍胜利的概率时,参赛对象m的表现特征对于比赛预测结果的贡献程度越高则表示其对自身所在第一参赛队伍作出的贡献越低),当参赛对象m属于第二参赛队伍时,则可以汇总得到参赛对象m在本场比赛中的贡献度评估结果是M2-M1。
具体的,结合上述可选实施例,在本公开实施例中,对于一个参赛对象而言,当模型输出的比赛预测结果表示该参赛对象自身所在参赛队伍获胜的概率时,则该参赛对象的表现特征对于比赛预测结果的贡献程度越高表示其对自身所在参赛队伍作出的贡献越高;当模型输出的比赛预测结果表示该参赛对象对手所在参赛队伍获胜的概率时,则该参赛对象的表现特征对于比赛预测结果的贡献程度越高则表示其对自身所在参赛队伍作出的贡献越低(即此时在进行汇总时,该参赛对象的多个表现特征的shapley值应该取其相反数)。
S103,根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
这里,贡献度评估结果的数值越大则表示参赛对象在本场比赛中的赛场表现(即上述表现特征)对于比赛结果预测模型输出的比赛预测结果的贡献程度越大,也即,贡献度评估结果的数值越大则表示参赛对象在本场比赛中对于获胜队伍取得胜利(即比赛结果)起到的作用越关键,因此,可以将上述贡献度评估结果的数字最大(或者上述贡献度评估结果的排名最靠前)的参赛对象确定为本场比赛的目标对象(如确定为本场比赛的MVP)。
具体的,作为一可选实施例,如果本场比赛的目标对象评选规则中没有限定目标对象必须出自获胜队伍(相当于在确定目标对象时,并不会考虑本场比赛的具体获胜队伍),则可以按照上述贡献度评估结果由高到低的顺序,对参与本场比赛的参赛对象进行排序,确定上述贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
具体的,作为一可选实施例,如果本场比赛的目标对象评选规则中要求从获胜队伍中选出目标对象(如比赛要求需要从获胜队伍中选择MVP选手),则在执行完上述步骤S102之后,还可以按照上述贡献度评估结果由高到低的顺序,对获胜队伍中的参赛对象进行排序,确定获胜队伍中贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
下面针对上述各步骤在本公开实施例中的具体实施过程,分别进行详细说明:
针对上述步骤S101中的比赛结果预测模型,图2示出了本公开实施例提供的一种训练得到比赛结果预测模型的方法的流程示意图,如图2所示,在执行步骤S101之前,所述方法包括如下步骤S201-S202,具体的:
S201,将多场比赛的比赛数据输入比赛结果预测模型中,通过所述比赛结果预测模型输出各场比赛的比赛预测结果。
这里,上述每场比赛的比赛数据中包括当场比赛中多个历史参赛对象的多个历史表现特征;其中,关于比赛数据以及表现特征的具体解释可以参考前述步骤S101处的相关说明内容,重复之处在此不再赘述。
这里,比赛结果预测模型可以是LightGBM模型,也可以是其他模型结构的分类模型,对于比赛结果预测模型所属的具体模型结构,本公开实施例不作任何限定。
需要说明的是,由于本公开是将shapley值的计算方式与比赛结果预测模型的比赛结果预测过程相结合,因此,考虑到树结构的比赛结果预测模型中可以使用treeshap算法更加快速地完成shapley值的计算,在本公开实施例中,可以优先选用树结构的比赛结果预测模型(例如上述LightGBM模型)。
S202,根据同场比赛的比赛预测结果与真实比赛结果之间的损失,对所述比赛结果预测模型进行训练,直至所述比赛结果预测模型达到收敛。
具体的,针对每场比赛,可以根据比赛结果预测模型输出的比赛预测结果与该场比赛的真实比赛结果之间的分类损失,对比赛结果预测模型进行训练,直至比赛结果预测模型达到收敛,得到包括调整好的模型参数在内的比赛结果预测模型。
针对上述步骤S101中表现特征的筛选标准,参考前述步骤S101处的相关说明内容可知,在执行步骤S101时,根据评选目标对象时需要参考的多种比赛特征,可以从该名参赛对象的赛场表现数据中,提取与每种比赛特征相匹配的目标赛场表现数据作为该名参赛对象在本场比赛中的多个表现特征;其中,为对上述评选目标对象时需要参考的多种比赛特征进行优化,图3示出了本公开实施例提供的一种优化评选目标对象时需要参考的多种比赛特征的方法的流程示意图,如图3所示,在执行步骤S101之前,所述方法包括如下步骤S301-S304,具体的:
S301,将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算同场历史比赛中多种比赛特征分别对应的shapley值。
这里,步骤S301中每种比赛特征对应的shapley值的具体计算方式可以参考前述步骤a中每种表现特征对应的shapley值的具体计算方式,重复之处在此不再赘述。
S302,根据同种比赛特征在多场历史比赛中对应的shapley值,计算该种比赛特征的平均shapley值。
具体的,以比赛特征y1为例,若共有m场历史比赛,则可以根据比赛特征y1在m场历史比赛中分别对应的m个shapley值,计算比赛特征y1在m场历史比赛中的平均shapley值(即上述m个shapley值的平均值)。
S303,根据多种比赛特征的平均shapley值,从多种比赛特征中,删除所述平均shapley值符合目标筛选条件的比赛特征,得到剩余的目标比赛特征。
这里,作为一可选实施例,符合目标筛选条件的比赛特征可以是:平均shapley值大于或者等于预设阈值的比赛特征。
这里,作为另一可选实施例,符合目标筛选条件的比赛特征还可以是:平均shapley值大于其他任意一种比赛特征且与其他任意一种比赛特征之间的平均shapley值的差距大于或者等于预设差距阈值(相当于在上述多种比赛特征中,平均shapley值远大于其他比赛特征的比赛特征)。
需要说明的是,参考上述两种可选实施例,需要删除的比赛特征(也即平均shapley值符合目标筛选条件的比赛特征)属于平均shapley值明显偏大的比赛特征,也即,计算出的平均shapley值可以表征一个比赛特征在上述多种比赛特征中的重要性值,当一个比赛特征的重要性值远大于其他比赛特征时,则基于该比赛特征可能会导致标签泄露,使得多种比赛特征对应的shapley值整体计算准确率降低的原因,可以删除该比赛特征,以便维持多种比赛特征对应的shapley值的整体计算准确率。
S304,从每个所述参赛对象在本场比赛的表现数据中,确定与所述目标比赛特征相匹配的表现数据作为所述参赛对象在本场比赛中的多个表现特征。
这里,步骤S304的具体实施方式可以参考前述步骤S101中关于比赛特征与表现特征的相关说明内容,重复之处在此不再赘述。
示例性的说明,以目标比赛特征包括累计获得的篮板球数量为例,则在执行步骤S304时,可以从每个参赛对象在本场比赛的表现数据中,确定每个参赛对象在本场比赛中累计获得的篮板球数量作为每个参赛对象在本场比赛中的一个表现特征。
在上述步骤S101-S103所示的目标对象确定方法的基础上,根据每个参赛对象在每场比赛中的贡献度评估结果,本公开实施例还提供了以下三种用于在赛季结算时确定本赛季的目标对象(如确定本赛季的MVP)的可选实施方式:
在一种可选的实施方式中,可以按照如下步骤b1-b2所述的方式确定本赛季的目标对象,具体的:
步骤b1、在赛季结算时,获取参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名作为第一平均贡献度评估结果排名。
这里,在步骤b1中,可以按照上述步骤S101-S103所示的目标对象确定方法,获取每个参赛对象在本赛季的每场比赛中的贡献度评估结果排名,然后,针对每个参赛对象,可以从获取到的每场比赛中的贡献度评估结果排名中,筛除出该参赛对象在获胜场次比赛中的多个贡献度评估结果排名,并计算出上述多个贡献度评估结果排名的平均值作为该参赛对象的第一平均贡献度评估结果排名。
步骤b2、从本赛季的参赛对象中,确定所述第一平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
这里,第一平均贡献度评估结果排名表示的是一个参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名,也即,第一平均贡献度评估结果排名越靠前(即数值越小)则表示参赛对象对于各获胜场次比赛中的比赛结果起到的贡献越大,越有资格成为本赛季的目标对象(如本赛季的MVP)。
在一种可选的实施方式中,可以按照如下步骤c1-c2所述的方式确定本赛季的目标对象,具体的:
步骤c1、在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果排名作为第二平均贡献度评估结果排名。
这里,步骤c1的具体实施方式可以参考前述步骤b1的具体实施方式,只是将步骤b1中的获胜场次替换成所有参与的比赛场次即可,重复之处在此不再赘述。
步骤c2、从本赛季的参赛对象中,确定所述第二平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
这里,第二平均贡献度评估结果排名表示的是一个参赛对象在本赛季各场次比赛中的平均贡献度评估结果排名,也即,第二平均贡献度评估结果排名越靠前(即数值越小)则表示参赛对象对于各场次比赛中的比赛结果起到的贡献越大,越有资格成为本赛季的目标对象(如本赛季的MVP)。
在一种可选的实施方式中,可以按照如下步骤d1-d2所述的方式确定本赛季的目标对象,具体的:
步骤d1、在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果。
这里,在步骤d1中,可以按照上述步骤S101-S103所示的目标对象确定方法,获取每个参赛对象在本赛季的每场比赛中的贡献度评估结果,然后,针对每个参赛对象,根据该参赛对象参加的比赛次数,计算出上述贡献度评估结果的平均值作为该参赛对象在本赛季各场比赛中的平均贡献度评估结果。
步骤d2、从本赛季的参赛对象中,确定所述平均贡献度评估结果最高的参赛对象作为本赛季的目标对象。
这里,平均贡献度评估结果表示的是一个参赛对象在本赛季各场次比赛中的平均贡献度评估结果,也即,平均贡献度评估结果的数值越大则表示参赛对象对于各场次比赛中的比赛结果起到的贡献越大,越有资格成为本赛季的目标对象(如本赛季的MVP)。
基于本公开实施例提供的上述目标对象的确定方法,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
基于同一发明构思,本公开还提供了与上述目标对象的确定方法对应的装置,由于本公开实施例中的目标对象的确定装置解决问题的原理与本公开实施例中上述目标对象的确定方法相似,因此目标对象的确定装置的实施可以参见上述目标对象的确定方法的实施,重复之处不再赘述。
参照图4所示,图4示出了本公开实施例所提供的一种目标对象的确定装置的结构示意图,其中,所述确定装置包括:
贡献分配模块401,被配置为根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
统计模块402,被配置为对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
评选模块403,被配置为根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
在一种可选的实施方式中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度时,贡献分配模块401,被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算所述表现特征的shapley值作为所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;其中,所述得分函数表征所述比赛结果预测模型中用于根据模型输入数据输出所述比赛预测结果的函数。
在一种可选的实施方式中,所述装置还包括特征筛选模块,其中,所述特征筛选模块被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算同场历史比赛中多种比赛特征分别对应的shapley值;
根据同种比赛特征在多场历史比赛中对应的shapley值,计算该种比赛特征的平均shapley值;
根据多种比赛特征的平均shapley值,从多种比赛特征中,删除所述平均shapley值符合目标筛选条件的比赛特征,得到剩余的目标比赛特征;
从每个所述参赛对象在本场比赛的表现数据中,确定与所述目标比赛特征相匹配的表现数据作为所述参赛对象在本场比赛中的多个表现特征。
在一种可选的实施方式中,所述装置还包括模型训练模块,其中,所述模型训练模块被配置为通过以下方法预先训练得到所述比赛结果预测模型:
将多场比赛的比赛数据输入比赛结果预测模型中,通过所述比赛结果预测模型输出各场比赛的比赛预测结果;其中,所述比赛数据中包括当场比赛中多个历史参赛对象的多个历史表现特征;
根据同场比赛的比赛预测结果与真实比赛结果之间的损失,对所述比赛结果预测模型进行训练,直至所述比赛结果预测模型达到收敛。
在一种可选的实施方式中,在所述根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象时,评选模块403,被配置为:
按照所述贡献度评估结果由高到低的顺序,对参与本场比赛的参赛对象进行排序,确定所述贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
在一种可选的实施方式中,所述装置还包括赛季结算模块,其中,所述赛季结算模块被配置为:
在赛季结算时,获取参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名作为第一平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第一平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,所述赛季结算模块还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果排名作为第二平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第二平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,所述赛季结算模块还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果;
从本赛季的参赛对象中,确定所述平均贡献度评估结果最高的参赛对象作为本赛季的目标对象。
基于本公开实施例提供的上述目标对象的确定装置,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
基于同一发明构思,本公开还提供了与上述目标对象的确定方法对应的电子设备,由于本公开实施例中的电子设备解决问题的原理与本公开实施例中上述目标对象的确定方法相似,因此电子设备的实施可以参见上述目标对象的确定方法的实施,重复之处不再赘述。
图5为本公开实施例提供的一种电子设备500的结构示意图,包括:处理器501、存储器502和总线503,存储器502存储有处理器501可执行的机器可读指令,当电子设备运行如实施例中的一种目标对象的确定方法时,处理器501与存储器502之间通过总线503通信,处理器501执行所述机器可读指令,处理器501执行所述机器可读指令时实现以下步骤,具体的:
根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
在一种可选的实施方式中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度时,处理器501被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算所述表现特征的shapley值作为所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;其中,所述得分函数表征所述比赛结果预测模型中用于根据模型输入数据输出所述比赛预测结果的函数。
在一种可选的实施方式中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度之前,处理器501被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算同场历史比赛中多种比赛特征分别对应的shapley值;
根据同种比赛特征在多场历史比赛中对应的shapley值,计算该种比赛特征的平均shapley值;
根据多种比赛特征的平均shapley值,从多种比赛特征中,删除所述平均shapley值符合目标筛选条件的比赛特征,得到剩余的目标比赛特征;
从每个所述参赛对象在本场比赛的表现数据中,确定与所述目标比赛特征相匹配的表现数据作为所述参赛对象在本场比赛中的多个表现特征。
在一种可选的实施方式中,处理器501被配置为通过以下方法预先训练得到所述比赛结果预测模型:
将多场比赛的比赛数据输入比赛结果预测模型中,通过所述比赛结果预测模型输出各场比赛的比赛预测结果;其中,所述比赛数据中包括当场比赛中多个历史参赛对象的多个历史表现特征;
根据同场比赛的比赛预测结果与真实比赛结果之间的损失,对所述比赛结果预测模型进行训练,直至所述比赛结果预测模型达到收敛。
在一种可选的实施方式中,在所述根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象时,处理器501被配置为:
按照所述贡献度评估结果由高到低的顺序,对参与本场比赛的参赛对象进行排序,确定所述贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
在一种可选的实施方式中,处理器501还被配置为:
在赛季结算时,获取参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名作为第一平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第一平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,处理器501还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果排名作为第二平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第二平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,处理器501还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果;
从本赛季的参赛对象中,确定所述平均贡献度评估结果最高的参赛对象作为本赛季的目标对象。
通过本公开实施例提供的上述电子设备,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
基于同一发明构思,本公开实施例还提供了一种计算机可读存储介质,该计算机可读存储介质上存储有计算机程序,该计算机程序被处理器运行时执行,所述处理器执行以下步骤:
根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
在一种可选的实施方式中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度时,所述处理器被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算所述表现特征的shapley值作为所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;其中,所述得分函数表征所述比赛结果预测模型中用于根据模型输入数据输出所述比赛预测结果的函数。
在一种可选的实施方式中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度之前,其中,所述处理器被配置为:
将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算同场历史比赛中多种比赛特征分别对应的shapley值;
根据同种比赛特征在多场历史比赛中对应的shapley值,计算该种比赛特征的平均shapley值;
根据多种比赛特征的平均shapley值,从多种比赛特征中,删除所述平均shapley值符合目标筛选条件的比赛特征,得到剩余的目标比赛特征;
从每个所述参赛对象在本场比赛的表现数据中,确定与所述目标比赛特征相匹配的表现数据作为所述参赛对象在本场比赛中的多个表现特征。
在一种可选的实施方式中,所述处理器被配置为通过以下方法预先训练得到所述比赛结果预测模型:
将多场比赛的比赛数据输入比赛结果预测模型中,通过所述比赛结果预测模型输出各场比赛的比赛预测结果;其中,所述比赛数据中包括当场比赛中多个历史参赛对象的多个历史表现特征;
根据同场比赛的比赛预测结果与真实比赛结果之间的损失,对所述比赛结果预测模型进行训练,直至所述比赛结果预测模型达到收敛。
在一种可选的实施方式中,在所述根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象时,所述处理器被配置为:
按照所述贡献度评估结果由高到低的顺序,对参与本场比赛的参赛对象进行排序,确定所述贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
在一种可选的实施方式中,所述处理器还被配置为:
在赛季结算时,获取参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名作为第一平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第一平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,所述处理器还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果排名作为第二平均贡献度评估结果排名;
从本赛季的参赛对象中,确定所述第二平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
在一种可选的实施方式中,所述处理器还被配置为:
在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果;
从本赛季的参赛对象中,确定所述平均贡献度评估结果最高的参赛对象作为本赛季的目标对象。
通过本公开实施例提供的上述计算机可读存储介质,根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含表现特征的条件下,表现特征对于比赛结果预测模型输出的比赛预测结果的贡献程度;对多个表现特征分别对应的贡献程度进行汇总,得到每个参赛对象在本场比赛中的贡献度评估结果;根据贡献度评估结果,从参赛对象中确定本场比赛的目标对象。这样本公开为目标对象的评选提供了更加公平客观且合理的依据,使得目标对象评选结果具有简单透明且可解释性强的特点,有利于帮助各种赛事主办方更加全面且客观地评估各个参赛对象的赛场表现。
在本公开实施例中,该计算机可读存储介质被处理器运行时还可以执行其它机器可读指令,以执行如实施例中其它所述的目标对象的确定方法,关于具体执行的目标对象的确定方法步骤和原理参见方法侧实施例的说明,在此不再赘述。
在本公开所提供的实施例中,应该理解到,所揭露系统和方法,可以通过其它的方式实现。以上所描述的系统实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,又例如,多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些通信接口,系统或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本公开提供的实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
应注意到:相似的标号和字母在下面的附图中表示类似项,因此,一旦某一项在一个附图中被定义,则在随后的附图中不需要对其进行进一步定义和解释,此外,术语“第一”、“第二”、“第三”等仅用于区分描述,而不能理解为指示或暗示相对重要性。
最后应说明的是:以上所述实施例,仅为本公开的具体实施方式,用以说明本公开的技术方案,而非对其限制,本公开的保护范围并不局限于此,尽管参照前述实施例对本公开进行了详细的说明,本领域的普通技术人员应当理解:任何熟悉本技术领域的技术人员在本公开揭露的技术范围内,其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化,或者对其中部分技术特征进行等同替换;而这些修改、变化或者替换,并不使相应技术方案的本质脱离本公开实施例技术方案的精神和范围。都应涵盖在本公开的保护范围之内。因此,本公开的保护范围应以所述权利要求的保护范围为准。

Claims (11)

  1. 一种目标对象的确定方法,所述确定方法包括:
    根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
    对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
    根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
  2. 根据权利要求1所述的确定方法,其中,所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度,包括:
    将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算所述表现特征的shapley值作为所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;其中,所述得分函数表征所述比赛结果预测模型中用于根据模型输入数据输出所述比赛预测结果的函数。
  3. 根据权利要求2所述的确定方法,其中,在所述根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度之前,所述确定方法还包括:
    将所述比赛结果预测模型中的得分函数作为shapley值计算式中的效用函数,计算同场历史比赛中多种比赛特征分别对应的shapley值;
    根据同种比赛特征在多场历史比赛中对应的shapley值,计算该种比赛特征的平均shapley值;
    根据多种比赛特征的平均shapley值,从多种比赛特征中,删除所述平均shapley值符合目标筛选条件的比赛特征,得到剩余的目标比赛特征;
    从每个所述参赛对象在本场比赛的表现数据中,确定与所述目标比赛特征相匹配的表现数据作为所述参赛对象在本场比赛中的多个表现特征。
  4. 根据权利要求1所述的确定方法,其中,通过以下方法预先训练得到所述比赛结果预测模型:
    将多场比赛的比赛数据输入比赛结果预测模型中,通过所述比赛结果预测模型输出各场比赛的比赛预测结果;其中,所述比赛数据中包括当场比赛中多个历史参赛对象的多个历史表现特征;
    根据同场比赛的比赛预测结果与真实比赛结果之间的损失,对所述比赛结果预测模型进行训练,直至所述比赛结果预测模型达到收敛。
  5. 根据权利要求1所述的确定方法,其中,所述根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象,包括:
    按照所述贡献度评估结果由高到低的顺序,对参与本场比赛的参赛对象进行排序,确定所述贡献度评估结果排名最高的参赛对象成为本场比赛的目标对象。
  6. 根据权利要求5所述的确定方法,其中,所述确定方法还包括:
    在赛季结算时,获取参赛对象在本赛季各获胜场次比赛中的平均贡献度评估结果排名作为第一平均贡献度评估结果排名;
    从本赛季的参赛对象中,确定所述第一平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
  7. 根据权利要求5所述的确定方法,其中,所述确定方法还包括:
    在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果排名作为第二平均贡献度评估结果排名;
    从本赛季的参赛对象中,确定所述第二平均贡献度评估结果排名最靠前的参赛对象作为本赛季的目标对象。
  8. 根据权利要求1所述的确定方法,其中,所述确定方法还包括:
    在赛季结算时,获取参赛对象在本赛季各场比赛中的平均贡献度评估结果;
    从本赛季的参赛对象中,确定所述平均贡献度评估结果最高的参赛对象作为本赛季的目标对象。
  9. 一种目标对象的确定装置,所述确定装置包括:
    贡献分配模块,被配置为根据每个参赛对象在本场比赛中的多个表现特征,分别确定在比赛结果预测模型的模型输入数据中包含所述表现特征的条件下,所述表现特征对于所述比赛结果预测模型输出的比赛预测结果的贡献程度;
    统计模块,被配置为对所述多个表现特征分别对应的所述贡献程度进行汇总,得到每个所述参赛对象在本场比赛中的贡献度评估结果;
    评选模块,被配置为根据所述贡献度评估结果,从所述参赛对象中确定本场比赛的目标对象。
  10. 一种电子设备,包括:处理器、存储器和总线,所述存储器存储有所述处理器可执行的机器可读指令,当电子设备运行时,所述处理器与所述存储器之间通过总线通信,所述机器可读指令被所述处理器执行时执行如权利要求1至8任一所述的目标对象的确定方法的步骤。
  11. 一种计算机可读存储介质,该计算机可读存储介质上存储有计算机程序,该计算机程序被处理器运行时执行如权利要求1至8任一所述的目标对象的确定方法的步骤。
PCT/CN2025/103408 2024-07-19 2025-06-25 一种目标对象的确定方法、装置、设备及存储介质 Pending WO2026016754A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202410979559.7 2024-07-19
CN202410979559.7A CN118846531A (zh) 2024-07-19 2024-07-19 一种目标对象的确定方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2026016754A1 true WO2026016754A1 (zh) 2026-01-22

Family

ID=93177101

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2025/103408 Pending WO2026016754A1 (zh) 2024-07-19 2025-06-25 一种目标对象的确定方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN118846531A (zh)
WO (1) WO2026016754A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118846531A (zh) * 2024-07-19 2024-10-29 网易(杭州)网络有限公司 一种目标对象的确定方法、装置、设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030073493A1 (en) * 2001-07-10 2003-04-17 Campaigne Philip James Method and system for real-time reportiing of team-member contributions to team achievement
US20050209717A1 (en) * 2004-03-08 2005-09-22 Flint Michael S Competitor evaluation method and apparatus
CN110147524A (zh) * 2019-05-10 2019-08-20 深圳市腾讯计算机系统有限公司 一种基于机器学习的比赛结果预测方法、装置及设备
CN113393063A (zh) * 2021-08-17 2021-09-14 深圳市信润富联数字科技有限公司 比赛结果预测方法、系统、程序产品及存储介质
CN116324668A (zh) * 2020-10-01 2023-06-23 斯塔特斯公司 从非职业跟踪数据预测nba天赋和质量
CN118846531A (zh) * 2024-07-19 2024-10-29 网易(杭州)网络有限公司 一种目标对象的确定方法、装置、设备及存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030073493A1 (en) * 2001-07-10 2003-04-17 Campaigne Philip James Method and system for real-time reportiing of team-member contributions to team achievement
US20050209717A1 (en) * 2004-03-08 2005-09-22 Flint Michael S Competitor evaluation method and apparatus
CN110147524A (zh) * 2019-05-10 2019-08-20 深圳市腾讯计算机系统有限公司 一种基于机器学习的比赛结果预测方法、装置及设备
CN116324668A (zh) * 2020-10-01 2023-06-23 斯塔特斯公司 从非职业跟踪数据预测nba天赋和质量
CN113393063A (zh) * 2021-08-17 2021-09-14 深圳市信润富联数字科技有限公司 比赛结果预测方法、系统、程序产品及存储介质
CN118846531A (zh) * 2024-07-19 2024-10-29 网易(杭州)网络有限公司 一种目标对象的确定方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN118846531A (zh) 2024-10-29

Similar Documents

Publication Publication Date Title
US12268963B2 (en) Game character behavior control method and apparatus, storage medium, and electronic device
WO2015103964A1 (en) Method, apparatus, and device for determining target user
Chen et al. Modeling intransitivity in matchup and comparison data
JP5874292B2 (ja) 情報処理装置、情報処理方法、及びプログラム
JP5879899B2 (ja) 情報処理装置、情報処理方法、及びプログラム
CN115600691B (zh) 联邦学习中的客户端选择方法、系统、装置和存储介质
CN110175299B (zh) 一种推荐信息确定的方法及服务器
CN107158708A (zh) 多玩家视频游戏匹配优化
CN108066989A (zh) 一种随机匹配组队方法、装置及应用服务器
CN110457534A (zh) 一种基于人工智能的数据处理方法、装置、终端及介质
CN108304853B (zh) 游戏相关度的获取方法、装置、存储介质和电子装置
CN111265878A (zh) 数据处理方法、装置、电子设备及存储介质
CN109011580A (zh) 残局牌面获取方法、装置、计算机设备及存储介质
CN114344916B (zh) 一种数据处理方法及相关装置
Lim et al. Revealing social identity phenomena in videogames with archetypal analysis
CN111905373A (zh) 一种基于博弈论和纳什均衡的人工智能决策方法及系统
CN113893547A (zh) 一种基于适应度函数的数据处理方法、系统及存储介质
CN113393063A (zh) 比赛结果预测方法、系统、程序产品及存储介质
JP6828834B2 (ja) 論理計算装置、論理計算方法、およびプログラム
CN118846531A (zh) 一种目标对象的确定方法、装置、设备及存储介质
CN118059504B (zh) 一种基于胜率图的竞赛匹配系统及方法
CN109544261A (zh) 一种基于扩散和数据质量的群智感知激励方法
CN114146401A (zh) 一种麻将智能决策方法、装置、存储介质及设备
CN113641856B (zh) 用于输出信息的方法和装置
JP7519199B2 (ja) 電子ゲーム情報処理装置及び電子ゲーム情報処理プログラム

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: 25839977

Country of ref document: EP

Kind code of ref document: A1