WO2013118493A1 - Dispositif de correction de condition de recommandation, procédé de correction de condition de recommandation et programme de correction de condition de recommandation - Google Patents

Dispositif de correction de condition de recommandation, procédé de correction de condition de recommandation et programme de correction de condition de recommandation Download PDF

Info

Publication number
WO2013118493A1
WO2013118493A1 PCT/JP2013/000626 JP2013000626W WO2013118493A1 WO 2013118493 A1 WO2013118493 A1 WO 2013118493A1 JP 2013000626 W JP2013000626 W JP 2013000626W WO 2013118493 A1 WO2013118493 A1 WO 2013118493A1
Authority
WO
WIPO (PCT)
Prior art keywords
context
time
recommendation
condition
user
Prior art date
Application number
PCT/JP2013/000626
Other languages
English (en)
Japanese (ja)
Inventor
健太郎 山崎
敏功 落合
佑嗣 小林
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 JP2013557426A priority Critical patent/JP5979159B2/ja
Publication of WO2013118493A1 publication Critical patent/WO2013118493A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising

Definitions

  • the present invention relates to a recommended condition correcting apparatus, a recommended condition correcting method, and a recommended condition correcting program, and in particular, a recommended condition correcting apparatus, a recommended condition correcting method, and a recommended condition correcting apparatus for correcting a recommended condition using a context of a recipient of recommendation information. , Related to the recommended condition correction program.
  • “Recommendation success rate” is an index of recommendation performance of the recommendation system 99.
  • the recommendation success rate is an index representing the accuracy of recommendation, and is a ratio at which the user receives a useful recommendation with respect to all the recommendations made to the user by the system.
  • Non-patent documents 1 and 2 show techniques for improving the recommendation success rate.
  • Non-Patent Document 1 uses a user's preference information represented by a user's purchase history and the like, and makes inferences using the purchased user's context, and improves the recommendation conditions to improve the recommendation success rate (purchasing recommended products). A technology for improving the rate).
  • Non-Patent Document 2 discloses a technique for improving a recommendation success rate by performing inference using a recommendation history including a user context at the time of recommendation and recommendation success / failure information and improving a recommendation condition.
  • Patent Documents 1 to 4 disclose a device for making a recommendation.
  • Non-Patent Documents 1 and 2 described above may not be able to improve the recommendation success rate. This problem occurs because the information used for inference for improving the recommendation success rate does not necessarily include the attribute information of the user that is the reason why the user uses the recommended content or service. That is, the information used for inference may not include information effective for inference for improving the recommendation success rate.
  • the attribute information of the user which is the reason for using the recommended content or service, is included in the context when the user decides to use the content or service, that is, the action decision time.
  • learning data is used for inference when receiving a recommendation, that is, when receiving a recommendation, when using a recommended content and service, that is, when a recommendation is successful. Context.
  • the above system can improve the recommendation success rate when the action decision is the same as when the recommendation is received or when the recommendation is successful, but when the action decision is different from the recommendation reception or when the recommendation is successful The recommendation success rate cannot be improved.
  • An object of the present invention is to provide a technique capable of improving the recommendation success rate even when the action decision is different from the recommendation reception or the recommendation success.
  • the recommendation condition correction apparatus includes a condition storage unit that stores the recommendation condition for transmitting recommendation information to the user when a context that is user status information matches the recommendation condition; Temporary storage means for storing a plurality of pairs of the user's context and its generation time, which are issued during a grace period from when the user receives the recommendation information to the time according to the recommendation information, and the temporary storage means Referring to FIG. 5, a decision-time determining unit that estimates a time point when the context starts to change and outputs a decision-time context that is the context at the time point, and a plurality of the decision-time contexts, Condition correcting means for correcting is provided.
  • the recommended condition correction program is a condition for storing, in a condition storage unit, the recommended condition for transmitting recommended information to the user when a context that is user state information matches the recommended condition.
  • a decision process that estimates a time point when the context starts to change with reference to the temporary storage unit, and outputs a context at the time of decision that is the context at the time point; and a plurality of decision times Based on the context, the computer executes a condition correction process for correcting the recommended condition.
  • the recommended condition correction method stores the recommended condition for transmitting recommendation information to the user in a condition storage unit when a context that is user status information matches the recommended condition, A plurality of pairs of the user's context and its occurrence time, which are issued during a grace period from when the user receives the recommendation information to the time according to the recommendation information, are stored in the temporary storage unit, and the temporary storage unit Referring to FIG. 6, the time when the context starts to change is estimated, the decision context that is the context at the time is output, and the recommendation condition is corrected based on a plurality of the decision contexts.
  • the present invention can improve the recommendation success rate even when the action decision is different from the recommendation reception or the recommendation success.
  • FIG. 1 It is a block diagram of the recommendation system 99 of 1st Embodiment.
  • An example of scenario 12 is shown.
  • An example of the context 13 is shown.
  • An example of the condition information 41 is shown.
  • An example of a plurality of condition information 41 stored in the condition storage unit 40 is shown.
  • An example of the timed context 81 stored in the temporary storage unit 80 is shown.
  • An example of the recommendation state 91 is shown.
  • An example of the timed context 81 created by the recommendation processing unit 95 is shown.
  • An example of the timed context 81 acquired from the temporary storage unit 80 is shown.
  • the frequency distribution table created from the context with time 81 acquired from the temporary storage unit 80 is shown.
  • An example of the context 31 at the time of decision determined based on the frequency distribution table is shown.
  • FIG. 31 An example of a decision context 31 stored in the decision context storage unit 30 is shown.
  • the seven decision contexts 31 acquired from the decision context storage unit 30 are shown.
  • the conditional expression of the recommended condition after correction is shown.
  • It is a scenario registration flowchart of 1st Embodiment.
  • It is a context processing flowchart of a 1st embodiment.
  • It is a success information processing flowchart of a 1st embodiment.
  • It is a flowchart of the creation and registration process of the context 31 at the time of determination which the determination part 70 at the time of determination performs.
  • It is a recommendation condition correction flowchart of a 1st embodiment It is a flowchart of the creation and registration process of the context 31 at the time of determination which the determination part 70 of a 2nd embodiment performs.
  • It is a block diagram of the recommendation condition recommendation apparatus 10 of 3rd Embodiment.
  • FIG. 1 is a configuration diagram of a recommendation system 99 according to the first embodiment of this invention.
  • the recommendation system 99 shown in FIG. 1 includes a recommendation condition correction device 10, an administrator terminal 11, a user terminal 15, a success information terminal 16, and a recommendation execution device 17.
  • the recommended condition correcting apparatus 10 includes a registration unit 60, a condition storage unit 40, a search unit 50, a recommendation processing unit 95, and a temporary storage unit 80, each of which includes a logic circuit and a storage device.
  • the recommendation state storage unit 90, the determination time determination unit 70, the determination time context storage unit 30, and the condition correction unit 20 are included.
  • the recommended condition correction device 10 may be a computer that operates under program control.
  • the registration unit 60, the search unit 50, the recommendation processing unit 95, the determination time determination unit 70, and the condition correction unit 20 are programs that are stored in a storage device by a processing device (processor) included in the computer. It may be realized by reading and executing.
  • the condition storage unit 40, the temporary storage unit 80, the recommended state storage unit 90, and the determination time context storage unit 30 may include a disk device provided in the computer.
  • the registration unit 60 receives the scenario 12 from the administrator terminal 11 and registers the condition information 41 in the condition storage unit 40.
  • FIG. 2 shows an example of the scenario 12.
  • the scenario 12 includes a scenario ID, recommendation conditions, and recommendation information.
  • the scenario ID is information for distinguishing the scenario 12.
  • the recommendation condition is information including a recommendation condition ID and a conditional expression.
  • the recommendation condition ID is information for distinguishing the recommendation condition.
  • the conditional expression expresses a condition for selecting a user to be recommended using part or all of the user context 13.
  • the context 13 is information indicating the state of the user and includes a plurality of attribute information.
  • the attribute information includes a pair of attribute name and attribute value.
  • the attribute name indicates the property of the attribute, and is, for example, a user ID, a position, and an age.
  • the attribute value indicates the value of the attribute, for example, Ichiro Tanaka, XX station, 30 years old.
  • FIG. 3 shows an example of the context 13.
  • This is a context 13 including the attribute information.
  • This context 13 indicates that “User 1 who is a man in his fifties is currently at the station, has 1000 steps, and is in good health”.
  • the conditional expression represents the condition of the context 13 to be satisfied by the user to be recommended by one or more attribute information, that is, one or more attribute name and attribute value pairs.
  • the attribute value may be a range of values.
  • Recommendation information is information related to recommendation.
  • the recommendation information may be a recommendation information ID indicating recommendation content, an address of a recommendation execution device 17 that makes a recommendation, an advertisement that is presented to a user that matches a conditional expression of recommendation conditions, and a program that is executed.
  • FIG. 2 shows a scenario 12 when the recommendation information is a recommendation information ID.
  • the condition information 41 is information including a condition ID, a scenario ID, a conditional expression, and association information.
  • FIG. 4 shows an example of the condition information 41.
  • the registration unit 60 creates the condition information 41 from the scenario 12.
  • the registration unit 60 acquires a recommendation condition ID, a scenario ID, a conditional expression, and recommendation information from the scenario 12, and sets the information as the condition ID, the scenario ID, the conditional expression, and the association information of the condition information 41.
  • the condition storage unit 40 stores one or more condition information 41.
  • FIG. 5 shows an example of a plurality of condition information 41 stored in the condition storage unit 40.
  • FIG. 5 shows an example in which the condition storage unit 40 stores two condition information 41 having a condition ID of recommendation condition 1 and recommendation condition 2.
  • the search unit 50 receives the context 13 from the user terminal 15, acquires the condition information 41 that matches the context 13 from the condition storage unit 40, and transmits the acquisition result to the recommendation processing unit 95.
  • a plurality of condition information 41 illustrated in FIG. 5 is stored in the condition storage unit 40, and the search unit 50 receives the context 13 illustrated in FIG.
  • the temporary storage unit 80 stores a timed context 81 from when a recommended user is recommended to when the recommendation is successful. For example, after receiving the recommendation information, the user terminal 15 transmits the user context 13 to the recommendation condition correction apparatus 10 every time it detects that the user context 13 has changed. The temporary storage unit 80 stores the context 13 transmitted from these user terminals 15.
  • FIG. 6 shows an example of the timed context 81 stored in the temporary storage unit 80.
  • the timed context 81 is information including the context 13, the time when the user is in the state of the context 13, that is, the time when the context 13 is generated, and the recommendation condition ID for which the user has been recommended. It is.
  • the time included in the context with time 81 may be an approximate value of the time when the user enters the context 13.
  • the time is, for example, the time when the user terminal 15 detects the user's context 13, the time when the user terminal 15 transmits the user's context 13 to the recommended condition correcting device 10, or the recommended condition correcting device 10 includes the timed context.
  • the time when 81 is stored in the temporary storage unit 80 may be used.
  • FIG. 6 shows a recommendation that user 1 receives recommendation content corresponding to recommendation condition 1 and five times after user 2 receives recommendation content corresponding to recommendation condition 1 and user 3 that recommends corresponding to recommendation condition 1.
  • FIG. 6 shows a recommendation that user 1 receives recommendation content corresponding to recommendation condition 1 and five times after user 2 receives recommendation content corresponding to recommendation condition 1 and user 3 that recommends corresponding to recommendation condition 1.
  • An example in which the context 13 changes once after receiving the contents is shown.
  • the recommended state storage unit 90 stores a recommended state 91.
  • the recommendation state 91 is information including a user ID and a recommendation condition ID.
  • the recommended state 91 stored in the recommended state storage unit 90 indicates that the user ID in the recommended state 91 has received the recommended content corresponding to the recommended condition ID.
  • FIG. 7 shows an example of the recommendation state 91.
  • FIG. 7 shows that user 1, user 2, and user 3 each received recommendation content corresponding to recommendation condition 1.
  • the recommendation processing unit 95 receives the context 13 and the condition information 41 from the search unit 50, acquires the recommended state 91 of the user from the recommended state storage unit 90, and based on the recommended state 91 and the context 13, the timed context 81 is created, and the created timed context 81 is registered in the temporary storage unit 80. At this time, the recommendation processing unit 95 creates a timed context 81 including the recommendation condition ID included in the user recommendation state 91 acquired from the recommendation state storage unit 90, the time when the context 13 is generated, and the received context 13. The time at which the context 13 occurs may be an approximate value as described above.
  • FIG. 8 shows an example of a timed context 81 created by the recommendation processing unit 95.
  • the recommendation processing unit 95 transmits a recommendation instruction including the condition information 41 and the user ID included in the context 13 to the recommendation execution device 17, and creates a recommendation state 91 of the user from the condition information 41, and recommends it. Register in the state storage unit 90.
  • the recommendation processing unit 95 creates a recommendation state 91 including a user ID included in the context 13 and a condition ID included in the condition information 41.
  • the determination time determination unit 70 receives the recommendation success information from the success information input terminal 16, deletes the recommended state 91 of the user from the recommendation state storage unit 90, and the time related to the user ID and the recommendation condition ID included in the recommendation success information
  • the attached context 81 is acquired from the temporary storage unit 80. What is acquired here is a timed context 81 that includes the context 13 that has occurred in the period from the recommendation time to the recommendation success time.
  • the determination time determination unit 70 determines the determination time context 31 from the timed context 81 and registers it in the determination time context storage unit 30.
  • Recommendation success information is information including a user ID and a recommendation condition ID.
  • the recommendation success information indicates that the user indicated by the user ID has taken an action capable of determining that the recommendation is successful for the recommendation content corresponding to the recommendation condition ID.
  • the recommendation success information indicates, for example, that a user who has received recommendation information recommending a certain product has purchased the product.
  • the recommendation success information is received from, for example, the success information input terminal 16 installed at the sales floor.
  • the decision determination unit 70 determines the context 13 at the time of action determination from the time distribution of the timed context 81 acquired from the temporary storage unit 80.
  • the decision determination unit 70 selects a context at the time when the context 13 starts to change from the timed contexts 81 from the recommendation time to the recommendation success time, and selects the decision context 31 from the selected context 13 and the condition ID. create.
  • the determination time determination unit 70 selects the context 13 at the time of action determination by one of the following methods.
  • the determination time determination unit 70 divides the period from the recommendation time to the recommendation success time into several sections, and a frequency distribution table in which the number of the timed contexts 81 including the time in each section is the frequency ( (Histogram).
  • the determination time determination unit 70 identifies the section with the highest frequency from the frequency distribution table, and the context 13 and the condition of the timed context 81 having the earliest time among the timed contexts 81 including the time in the section.
  • the ID is the context 31 at the time of decision.
  • the determination time determination unit 70 may use a period from the earliest time to the latest time that appears in the acquired timed context 81 as the period from the recommendation time to the recommendation success time.
  • the determination time determination unit 70 creates the same frequency distribution table as in Method 1 above.
  • the determination time determination unit 70 identifies a section immediately preceding the section with the highest frequency from the frequency distribution table, and the timed context having the latest time among the timed contexts 81 including the time in the section.
  • the context 13 of 81 and the condition ID are set as the context 31 at the time of decision.
  • FIG. 9 shows an example of a timed context 81 acquired from the temporary storage unit 80.
  • FIG. 10 shows a frequency distribution table created from the timed context 81.
  • FIG. 11 shows an example of the decision context 31 determined based on the frequency distribution table.
  • FIG. 10 shows a frequency distribution table when the time interval appearing in the timed context 81 is divided into four intervals.
  • the decision context storage unit 30 stores the decision context 31.
  • the decision context 31 is information including the condition ID and the context 13.
  • the context 13 included in the context 31 at the time of determination is the context 13 at the time when the user who received the recommendation of the service and the content decides to use the service and the content.
  • FIG. 12 shows an example of the decision context 31 stored in the decision context storage unit 30. This example shows that the recommended service and content are used nine times, and the context 13 when the user decides to use the recommended service and content in each usage.
  • condition correction unit 20 When the condition correction unit 20 receives the recommendation condition correction request including the recommendation condition ID, the condition correction unit 20 refers to the decision context 31 stored in the decision context storage unit 30 and refers to the recommendation stored in the condition storage unit 40. Correct the conditional expression for the condition ID.
  • the recommendation condition modification request includes a recommendation condition ID and an attribute information addition threshold.
  • the attribute information addition threshold is a lower limit value for adding attribute information to the recommendation condition when the success contribution rate of each attribute information is calculated from the context 31 at the time of decision.
  • the condition correction unit 20 adds attribute information whose success contribution rate is equal to or higher than the attribute information deletion threshold to the recommendation condition.
  • the condition modification unit 20 When receiving the recommended condition modification request, the condition modification unit 20 creates a condition information 41 acquisition request and a decision context 31 acquisition request.
  • the condition information 41 acquisition request and the decision context 31 acquisition request include a recommendation condition ID.
  • the condition correction unit 20 transmits a condition information 41 acquisition request to the condition storage unit 40, and acquires the condition information 41 as a response.
  • the condition correction unit 20 transmits a determination context 31 acquisition request to the determination context storage unit 30 and receives the determination context 31 as a response.
  • N the number of received contexts 31 at the time of decision is N (1 or more). Then, the condition correction unit 20 calculates the success contribution ratio of the attribute information using the decision context 31 and corrects the conditional expression of the condition information 41.
  • the condition correction unit 20 calculates the success contribution ratio and corrects the conditional expression by a method using a general probability model, a method using a nearest neighbor method, or other methods. For example, the condition correction unit 20 calculates the success contribution rate using the following formula for all attribute information except the attribute information whose attribute is the user ID.
  • Success contribution rate the number of contexts 31 at the time of decision when the attribute is the attribute value / N
  • the condition correction unit 20 corrects the conditional expression based on the following two criteria. 1) Among the attribute information of attributes not included in the conditional expression, the attribute having the highest success contribution rate and not less than the attribute information addition threshold is added to the conditional expression. 2) Change an attribute value for an attribute that is already included in the conditional expression and has a higher success contribution rate when the attribute value is changed.
  • condition correction unit 20 includes the seven decision contexts 31 shown in FIG. Is obtained from the context storage unit 30 at the time of decision.
  • the attribute information addition threshold is, for example, 5/7 (about 71%).
  • the attributes that are the targets of 1) are “gender” and “step count”.
  • FIG. 14 shows a conditional expression of the recommended condition after correction.
  • the administrator terminal 11 is an information processing apparatus including an input device such as a keyboard and a mouse, an output device such as a liquid crystal display, a processing device that operates under program control, and a storage device including a memory.
  • the administrator terminal 11 creates a scenario 12 and transmits it to the recommended condition correction device 10. Further, the administrator terminal 11 creates a recommended condition correction request 14 and transmits it to the recommended condition correction apparatus 10.
  • the user terminal 15 is an information processing apparatus including an input device such as a button, a GPS (Global Positioning System) sensor, an acceleration sensor, and a microphone, an output device such as a liquid crystal display, a processing device, and a storage device such as a memory. is there.
  • an input device such as a button, a GPS (Global Positioning System) sensor, an acceleration sensor, and a microphone
  • an output device such as a liquid crystal display
  • a processing device such as a memory.
  • the user terminal 15 transmits a context 13 to the recommended condition correction device 10 whenever there is a change in the value of the input device.
  • the context 13 is a value such as a position acquired by the sensor, a health state input from the input device, or a gender stored in the storage device of the user terminal 15.
  • a part of the context 13 such as gender may be stored in the search unit 50 or the like as user profile information in association with the user ID or the like.
  • the search unit 50 completes the context 13 by adding the stored attribute information to the attribute information received from the user terminal 15.
  • the user terminal 15 may transmit the context 13 to the recommended condition correction device 10 at regular intervals.
  • the search unit 50 may detect a change in the context 13.
  • the success information input terminal 16 is an information processing apparatus including an input device such as a keyboard and a mouse, an output device such as a liquid crystal display, a processing device operated by program control, and a storage device including a memory. Is transmitted to the recommended condition correction apparatus 10.
  • the success information input terminal 16 is, for example, a POS terminal (Point) of Sales) installed in a store that sells products recommended by the recommendation information.
  • the success information input terminal 16 can receive the context 13 from the user terminal 15 possessed by the user who visited the store, for example, by proximity communication.
  • the success information input terminal 16 stores, for example, correspondence information between sold product codes and recommendation condition IDs, and may create success information when inputting the sales of products.
  • the recommendation execution device 17 is an information processing device including an input device such as a keyboard and a mouse, an output device such as a liquid crystal display, a processing device operated by program control, and a storage device including a memory.
  • the recommendation execution device 17 receives information including the recommendation condition and the recommendation content, and transmits the recommendation information to the user terminal 15.
  • the operation of the recommendation system 99 according to the present embodiment is roughly divided into a scenario registration flow for registering the scenario 12, a context processing flow performed every time the context 13 is received, and a successful information processing flow performed every time the success information is received. And a recommended condition correction flow for correcting the recommended conditions.
  • FIG. 15 is a scenario registration flowchart showing the operation of the system according to the first embodiment.
  • the registration unit 60 receives the scenario 12 and proceeds to S12.
  • the registration unit 60 acquires the scenario ID, the recommended condition ID, the conditional expression, and the recommended information from the scenario 12, and sets the condition information 41 using the respective information as the scenario ID, the conditional ID, the conditional expression, and the association information. Create and proceed to S13.
  • the registration unit 60 transmits the condition information 41 to the condition storage unit 40, and the condition storage unit 40 stores the condition information 41 and ends the scenario registration flow.
  • FIG. 16 is a context processing flowchart showing the operation of the system according to the first embodiment.
  • the search unit 50 receives the context 13 and proceeds to S22.
  • the search unit 50 searches the condition storage unit 40 for the condition information 41 that matches the context 13 received in S21, and transmits the condition information 41 and the context 13 that are the search results to the recommendation processing unit 95. Proceed to S23. When the matching condition information 41 is not found, the search unit 50 transmits that the search result does not match.
  • the recommendation processing unit 95 proceeds to S24 if the context 13 matches one or more condition information 41 as a result of the search in S22 (Y in S23), and if there is no match (S23 N) and proceed to S26.
  • the recommendation processing unit 95 creates a recommendation state 91 including a combination of the user ID included in the received context 13 and the recommendation condition ID included in the condition information 41, and transmits the recommendation state 91 to the recommendation state storage unit 90.
  • the recommendation state storage unit 90 stores the received recommendation state 91 and proceeds to S25.
  • the recommendation processing unit 95 transmits a recommendation instruction including the user ID and the condition information 41 included in the received context 13 to the recommendation execution device 17, and the process proceeds to S26.
  • the recommendation execution apparatus device 17 transmits the recommendation information to the user terminal 15 indicated by the user ID.
  • the recommendation processing unit 95 acquires the recommendation state 91 including the user ID included in the context 13 received in the recommendation state storage unit 90. If the recommendation state 91 exists (Y in S26), the recommendation processing unit 95 proceeds to S27, and if it does not exist (N in S26), the process ends.
  • the recommendation processing unit 95 creates a timed context 81 including the recommendation condition ID, the current time, and the context 13 included in the recommendation state 91 acquired in S26, and transmits it to the temporary storage unit 80.
  • the temporary storage unit 80 stores the received context with time 81 and ends the process.
  • FIG. 17 is a success information processing flowchart showing the operation of the system according to the first embodiment.
  • the determination time determination unit 70 receives the success information including the user ID and the recommendation condition ID, and proceeds to S32.
  • the determination time determination unit 70 acquires a recommendation state 91 having a user ID and a recommendation condition ID included in the success information from the recommendation state storage unit 90.
  • the determination unit 70 proceeds to S33, and when it does not exist (N in S32), the process ends.
  • the determination time determination unit 70 deletes the recommendation state 91 having the user ID and the recommendation condition ID included in the success information from the recommendation state storage unit 90, and proceeds to S34.
  • the determination time determination unit 70 acquires the timed context 81 having the user ID and the recommendation condition ID included in the success information from the temporary storage unit 80, and proceeds to S35. In S35, the determination time determination unit 70 proceeds to S36 if one or more timed contexts 81 acquired in S34 exist (Y in S35), and proceeds to S36 if it does not exist (N in S35). Exit.
  • the determination time decision unit 70 creates the decision time context 31, and transmits the created decision time context 31 to the decision time context storage unit 30.
  • the decision-time context storage unit 30 stores the received decision-time context 31 and ends the process.
  • FIG. 18 is a flowchart of the creation and registration process of the decision context 31 executed by the decision determination unit 70.
  • the determination time determination unit 70 creates a time interval in which the interval from the earliest time to the latest time included in the timed context 81 acquired from the temporary storage unit 80 is divided into a plurality of intervals, and in S42 move on.
  • a value preset in the system may be used, or a value calculated from the number of contexts with time 81 may be used.
  • the determination time determination unit 70 may divide the time between the time when the recommendation execution device 17 makes the recommendation and the time when the success information is input from the success information input terminal 16 into sections.
  • the determination time determination unit 70 creates the frequency distribution table by counting the number of the timed contexts 81 included in each section based on the time of the timed context 81, and proceeds to S43. In S43, the determination time determination unit 70 acquires the time interval in which the frequency is maximum from the frequency distribution table, and proceeds to S44.
  • the determination time determination unit 70 selects the timed context 81 with the earliest time among the timed contexts 81 in the time interval with the maximum frequency, and proceeds to S45.
  • the determination time determination unit 70 may select the context 13 of the timed context 81 of the latest time among the timed contexts 81 of the time interval immediately before the time interval with the highest frequency.
  • the decision determination unit 70 creates the context 13 of the timed context 81 selected in S44 and the decision context 31 including the received recommendation condition ID, and decides the created decision context 31.
  • the decision-time context storage unit 30 stores the received decision-time context 31 and ends the process.
  • FIG. 19 is a recommended condition correction flowchart showing the operation of the system according to the first embodiment.
  • the condition correction unit 20 receives the recommended condition correction request, and proceeds to S52.
  • the recommendation condition correction request includes a recommendation condition ID.
  • the condition correction unit 20 creates a condition information 41 acquisition request including the recommended condition ID and transmits it to the condition storage unit 40.
  • the condition storage unit 40 receives the condition acquisition request, the condition storage unit 40 returns condition information 41 having a recommended condition ID that matches the recommended condition ID included in the condition acquisition request to the condition correction unit 20. If there is no condition information 41 having a matching recommended condition ID, the condition storage unit 40 returns “no match”.
  • the condition correction unit 20 receives the condition information 41 or “no match” from the condition storage unit 40, and proceeds to S53. In S53, the condition correction unit 20 proceeds to S54 if there is one or more condition information 41 matched in S52 (Y in S53), and ends the process in the case of “No match” (N in S53). To do.
  • the condition correction unit 20 creates a decision-time context 31 acquisition request including the recommended condition ID included in the condition acquisition request, and transmits it to the decision-time context storage unit 30.
  • the decision context storage unit 30 receives the decision context 31 acquisition request, the decision context storage unit 30 transmits the decision context 31 including the recommended condition ID to the condition correction unit 20.
  • the condition correction unit 20 acquires the context 31 at the time of determination, and proceeds to S55.
  • condition correction unit 20 proceeds to S56, and if none exists (N in S55), the process is terminated. .
  • condition correction unit 20 calculates the success contribution rate of the attribute information using the decision context 31 acquired in S54, corrects the conditional expression of the condition information 41, and ends the process.
  • the recommendation condition ID is not necessary for the context with time 81, the recommendation state 91, the context 31 at the time of decision, and the related conditions. No processing is necessary. Further, in the recommendation system 99, when there is only one recommendation condition for each user, the user ID is not necessary for the timed context 81, the recommendation state 91, the context 13, and the like. No processing is necessary. Further, when both the user and the recommendation condition are independent, neither the recommendation condition ID nor the user ID is required.
  • the recommendation condition correcting apparatus 10 can improve the recommendation success rate even when the user's action decision is different from the recommendation reception or the recommendation success.
  • the reason is that the decision determination unit 70 estimates the time point when the user decides the action from the context 13 from the user recommendation time to the action time, and outputs the context 13 at that time as the decision context 31. It is. Furthermore, the condition correction unit 20 improves the recommendation condition using the context 31 at the time of decision as learning data.
  • the recommended condition correcting apparatus 10 can appropriately estimate the context 13 at the time of user's action decision.
  • the reason is that the determination unit 70 determines a time interval in which the user context 13 frequently changes, and estimates the context 13 at the time of user action determination based on the interval.
  • the block diagram of the second embodiment is the same as the block diagram of the first embodiment.
  • the second embodiment of the present invention is different from the first embodiment in the method of selecting the context 13 of the decision context 31 performed by the decision determination unit 70.
  • the determination time determination unit 70 creates a frequency distribution table of the number of contexts 81 with time as in the first embodiment. Thereafter, the determination time determination unit 70 looks at the frequency from the latest time interval toward the earliest time, and has the earliest time among the timed contexts 81 in the interval after the interval where the frequency becomes 0 for the first time.
  • the context 13 of the timed context 81 is selected as the context 31 at the time of decision.
  • the operation of the recommendation system 99 according to the present embodiment is roughly divided into a scenario registration flow for registering the scenario 12, a context processing flow performed every time the context 13 is received, and a successful information processing flow performed every time success information is received.
  • the recommendation condition correction flow for correcting the recommendation condition can be divided.
  • the scenario registration flow, context processing flow, and recommendation condition correction flow are the same as those in the first embodiment.
  • the flow of the second embodiment is different from the first embodiment in the flow of creating and registering the context 31 at the time of decision shown in FIG. 18 in the successful information processing flow.
  • FIG. 20 is a flowchart of the creation and registration process of the decision time context 31 executed by the decision time decision unit 70 in the second embodiment.
  • the determination time determination unit 70 creates a plurality of time intervals, and the number of timed contexts 81 included in each interval based on the time of the timed context 81, as in the first embodiment. To create a frequency distribution table, and proceed to S63.
  • the determination time determination unit 70 acquires an unprocessed time section and its frequency from the later time side of the frequency distribution table, and proceeds to S64.
  • S64 if the frequency of the time interval acquired in S63 is 0 (Y in S64), the determination time determination unit 70 proceeds to S65, and if it is not 0 (N in S64), the time interval has been processed as S63. Proceed to
  • the determination time determination unit 70 processes the time interval processed immediately before the time interval where the frequency becomes 0 in S64, that is, the time interval one time later than the time interval where the frequency becomes 0.
  • the timed context 81 having the earliest time is selected from the timed contexts 81, and the process proceeds to S66.
  • the decision-time determination unit 70 creates the context 13 for decision including the context 13 of the selected timed context 81 and the received recommendation condition ID, and uses the created decision-time context 31 as the context for decision.
  • the data is transmitted to the storage unit 30.
  • the decision-time context storage unit 30 stores the received decision-time context 31 and ends the process.
  • the recommended condition correction apparatus 10 can appropriately estimate the context 13 at the time of user's action decision.
  • the reason is that the determination unit 70 determines a time interval in which the user context 13 starts to change, and estimates the context 13 at the time of the user's action determination based on the interval.
  • the recommended condition correction apparatus 10 includes a condition storage unit 40, a temporary storage unit 80, a determination time determination unit 70, and a condition correction unit 20.
  • the condition storage unit 40 stores a recommendation condition for transmitting recommendation information to the user when the context 13 as the user state information matches the recommendation condition.
  • Temporary storage unit 80 stores a plurality of pairs of the user's context 13 and its occurrence time, which are issued during a grace period that is a period from when the user receives the recommendation information until the user follows the recommendation information.
  • the decision time determination unit 70 refers to the temporary storage unit 80, estimates the time when the context 13 starts to change, and outputs the context 13 at that time.
  • the condition correcting unit 20 corrects the recommended condition based on the context 13 at the time of a plurality of decisions.
  • the recommendation condition correction apparatus 10 can improve the recommendation success rate even when the user's action decision is different from the recommendation reception time or the recommendation success time.
  • the reason is that the determination time determination unit 70 estimates the time point when the user has determined the action from the context 13 from the user recommendation time to the action time, and outputs the context 13 at that time. Furthermore, the condition correction unit 20 improves the recommendation condition using the context 13 at the time of decision as learning data.
  • the present invention can be applied to a recommendation system 99 represented by an advertisement distribution system, a control system, a notification system, an expert system, a navigation system, and the like. Furthermore, the present invention can be applied to the evaluation and optimization of a group or rule of a system that executes specific processing at high speed for input evaluation target information such as grouping processing, stream processing, and wrinkle rule matching.

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Information Transfer Between Computers (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

Un système de recommandation classique ne peut pas améliorer le taux de réussite des recommandations lorsque le moment de la décision d'action par l'utilisateur est différent du moment où la recommandation est reçue ou différent du moment où la recommandation est une réussite. Un dispositif de correction de condition de recommandation est doté : d'un moyen de stockage de conditions destiné à stocker les conditions de recommandation afin de transmettre les informations recommandées à l'utilisateur lorsque le contexte, qui constitue les informations d'état de l'utilisateur, correspond à la condition de recommandation ; d'un moyen de stockage temporaire conçu pour stocker une pluralité de paires composées d'une part du contexte utilisateur créé au cours de la période d'attente, c'est-à-dire la période avant que les informations de recommandation reçues par l'utilisateur soient conformes aux informations de recommandation, et d'autre part du moment de génération ; d'un moyen de détermination de moment de décision qui consulte ledit moyen de stockage temporaire, estime le moment où le contexte a commencé à changer, et délivre le contexte du moment de décision, c'est-à-dire le contexte à ce moment de décision ; et d'un moyen de correction de condition prévu pour corriger la condition de recommandation sur la base de la pluralité de contextes de moment de décision.
PCT/JP2013/000626 2012-02-09 2013-02-06 Dispositif de correction de condition de recommandation, procédé de correction de condition de recommandation et programme de correction de condition de recommandation WO2013118493A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2013557426A JP5979159B2 (ja) 2012-02-09 2013-02-06 推薦条件修正装置、推薦条件修正方法、および、推薦条件修正プログラム

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2012-026280 2012-02-09
JP2012026280 2012-02-09

Publications (1)

Publication Number Publication Date
WO2013118493A1 true WO2013118493A1 (fr) 2013-08-15

Family

ID=48947263

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2013/000626 WO2013118493A1 (fr) 2012-02-09 2013-02-06 Dispositif de correction de condition de recommandation, procédé de correction de condition de recommandation et programme de correction de condition de recommandation

Country Status (2)

Country Link
JP (1) JP5979159B2 (fr)
WO (1) WO2013118493A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015056360A1 (fr) * 2013-10-16 2015-04-23 カルチュア・コンビニエンス・クラブ株式会社 Système d'analyse/évaluation de données de clientèle
CN109983458A (zh) * 2017-09-18 2019-07-05 华为技术有限公司 一种推荐方法及终端

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005015902A1 (fr) * 2003-08-06 2005-02-17 Matsushita Electric Industrial Co., Ltd. Appareil de recommandation de programmes
JP2007323233A (ja) * 2006-05-31 2007-12-13 National Institute Of Information & Communication Technology 対話ロボットを用いた理由説明サービス処理方法およびその装置，およびそのプログラム
WO2008126775A1 (fr) * 2007-04-05 2008-10-23 Nec Corporation Appareil, système et procédé de recommandation d'informations, système de recommandation d'informations et programme de recommandation d'informations

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9043296B2 (en) * 2010-07-30 2015-05-26 Microsoft Technology Licensing, Llc System of providing suggestions based on accessible and contextual information

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005015902A1 (fr) * 2003-08-06 2005-02-17 Matsushita Electric Industrial Co., Ltd. Appareil de recommandation de programmes
JP2007323233A (ja) * 2006-05-31 2007-12-13 National Institute Of Information & Communication Technology 対話ロボットを用いた理由説明サービス処理方法およびその装置，およびそのプログラム
WO2008126775A1 (fr) * 2007-04-05 2008-10-23 Nec Corporation Appareil, système et procédé de recommandation d'informations, système de recommandation d'informations et programme de recommandation d'informations

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
KOJI ITO ET AL.: "A Study of the user modeling method for personal life assistance service", IEICE TECHNICAL REPORT, vol. 109, no. 272, 5 November 2009 (2009-11-05), pages 121 - 128 *
TAKANOBU SOMA ET AL.: "End-User Modification of Conditions for Activating Home Context-Aware Services", IPSJ SIG NOTES, vol. 2008, no. 110, 6 November 2008 (2008-11-06), pages 1 - 7 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015056360A1 (fr) * 2013-10-16 2015-04-23 カルチュア・コンビニエンス・クラブ株式会社 Système d'analyse/évaluation de données de clientèle
JP2015079331A (ja) * 2013-10-16 2015-04-23 カルチュア・コンビニエンス・クラブ株式会社 顧客データ分析・検証システム
CN104813315A (zh) * 2013-10-16 2015-07-29 文化便利俱乐部株式会社 顾客数据分析/验证系统
CN109983458A (zh) * 2017-09-18 2019-07-05 华为技术有限公司 一种推荐方法及终端
CN109983458B (zh) * 2017-09-18 2024-02-09 华为技术有限公司 一种推荐方法及终端

Also Published As

Publication number Publication date
JP5979159B2 (ja) 2016-08-24
JPWO2013118493A1 (ja) 2015-05-11

Similar Documents

Publication Publication Date Title
CN105447730B (zh) 目标用户定向方法及装置
US20200027033A1 (en) Updating Machine Learning Models On Edge Servers
US11983646B2 (en) Bias scoring of machine learning project data
US20180342004A1 (en) Cumulative success-based recommendations for repeat users
US10474957B2 (en) Method and apparatus for forecasting characteristic information change
JP5745993B2 (ja) 地域基盤の検索語を推薦するシステム及びその方法
US10162868B1 (en) Data mining system for assessing pairwise item similarity
US10552863B1 (en) Machine learning approach for causal effect estimation
US9710480B2 (en) Method and apparatus for performing an incremental update of a recommendation model
KR20170137129A (ko) 디바이스가 링크될 확률에 기초하여 디바이스의 컨텐츠를 선택하기 위한 시스템 및 방법
WO2010010654A1 (fr) Dispositif d'estimation d'utilisation
JP2015526795A5 (fr)
CN110033097B (zh) 基于多个数据域确定用户与物品的关联关系的方法及装置
US20200098031A1 (en) Product recommending apparatus and non-transitory computer readable medium
JP2010079608A (ja) 情報配信システム、情報配信サーバ、情報配信方法および情報配信プログラム
JP2021508395A (ja) パーソナライズされた推奨を生成するために適合されたクライアント、サーバ、およびクライアント−サーバシステム
US11854051B2 (en) Systems and methods for generating efficient iterative recommendation structures
US20130031093A1 (en) Information processing system, information processing method, program, and non-transitory information storage medium
US20150169794A1 (en) Updating location relevant user behavior statistics from classification errors
WO2013088682A1 (fr) Dispositif, procédé et programme de correction de conditions de recommandation
JP5979159B2 (ja) 推薦条件修正装置、推薦条件修正方法、および、推薦条件修正プログラム
JP5504220B2 (ja) 提供情報選択装置、方法及びプログラム
CN109658187A (zh) 推荐云服务商的方法、装置、存储介质及电子设备
US20210241171A1 (en) Machine learning feature engineering
JP2014074961A (ja) 商品推薦装置及び方法及びプログラム

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: 13746099

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2013557426

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 13746099

Country of ref document: EP

Kind code of ref document: A1