JP5622880B2 - Item recommendation system, item recommendation method, and item recommendation program - Google Patents

Description

  The present invention relates to an item recommendation system for recommending items to a user, an item recommendation method, and an item recommendation program.

  A number of item recommendation methods have been proposed. In any method, when creating a recommendation list, it is necessary to calculate a recommendation score for a combination of a user and an item.

  For example, in the technique described in Non-Patent Document 1, the purchase probability of each item of each user is calculated using the purchase history (information on which user purchased which item), and the item is assigned to each user in descending order of the value. Present a list of In the recommendation system based on this technology, the purchase probability of all items is calculated for each user.

  In Patent Literature 1, the user is allowed to specify the text of the part of interest, the user's interest is estimated based on the appearance frequency of the words included therein, the item is searched based on the word representing the user's interest, Present items that the user is interested in. In this case, it is necessary to calculate the degree of coincidence between the interest and the item for all items.

JP 11-338869 A

Resnick, P. et al., “GroupLens: An open architecture for collaborative filtering of netnews”. CSCW '94: Proceedings of the 1994 ACM conference on Computer supported cooperative work, ACM Press New York, 1994.

  As in the background art, when calculating a recommendation score for a combination of a user and an item for creating a recommendation list, the calculation cost increases as the number of items increases.

  Specifically, when an item is recommended using the method of the above-described Patent Document 1, recommendation scores corresponding to the number of combinations of the keyword of interest and the item are calculated. When an item is recommended using the method of Non-Patent Document 1, a recommendation score for the number of users and items is calculated.

  The present invention has been made in view of the above points. An item recommendation system, an item recommendation method, and an item recommendation program for reducing the calculation cost and processing load in item recommendation by narrowing down the number of items for calculating a recommendation score. The purpose is to provide.

In order to achieve the above object, the present invention according to claim 1 is an item recommendation system,
Purchase information acquisition means for acquiring purchase information for each item and storing it in the purchase information storage means;
The classification of each purchase information stored in the purchase information storage means is determined, the item selection rate of each item is calculated for each classification using the determined classification of the purchase information, and stored in the item selection rate storage means Item selection rate calculation means,
Using the item selection rate of each item calculated for each category, for each category, an item with a high item selection rate among the items belonging to each category is extracted as a candidate item, and extracted for each category. Candidate item determining means for storing candidate items in the candidate item storage means;
For each candidate item stored in the candidate item storage means, a recommendation score is calculated for each user using the item selection rate and the purchase information stored in the purchase information storage unit, and the user is based on the recommendation score. The gist of the present invention is to include a recommendation list creating unit that selects a recommended item for each recommendation, generates a recommendation list, and stores the recommendation list in the recommendation list storage unit.

  The present invention according to claim 2 is an item recommendation system, wherein when an item recommendation request is received from a terminal, a recommendation list corresponding to a user specified in the item recommendation request is extracted from the recommendation list storage unit, The gist of the invention is to further include recommendation processing means to be presented to the terminal.

The present invention according to claim 3 is an item recommendation method performed by a computer,
A purchase information acquisition step of acquiring purchase information for each item and storing it in a purchase information storage unit;
The classification of each purchase information stored in the purchase information storage unit is determined, the item selection rate of each item is calculated for each category using the determined classification of the purchase information, and stored in the item selection rate storage unit An item selection rate calculation step;
Using the item selection rate of each item calculated for each category, for each category, an item with a high item selection rate among the items belonging to each category is extracted as a candidate item, and extracted for each category. A candidate item determination step of storing the candidate item in the candidate item storage unit;
For each candidate item stored in the candidate item storage unit, a recommendation score is calculated for each user using the item selection rate and the purchase information stored in the purchase information storage unit, and the user is based on the recommendation score. The gist is to perform a recommendation list creation step of generating a recommendation list by selecting a recommended item for each recommendation and storing the recommendation list in a recommendation list storage unit.

  The present invention according to claim 4 is an item recommendation method, wherein when an item recommendation request is received from a terminal, a recommendation list corresponding to a user specified by the item recommendation request is extracted from the recommendation list storage unit, The gist is to further perform the recommendation processing step presented to the terminal.

  The gist of the present invention described in claim 5 is an item recommendation program for causing a computer to function as the item recommendation system.

  According to the present invention, it is possible to provide an item recommendation system, an item recommendation method, and an item recommendation program that reduce the number of items for calculating a recommendation score and reduce the calculation cost and processing load in item recommendation.

It is a lineblock diagram of an item recommendation system concerning an embodiment of the present invention. It is a figure which shows an example of a purchase information table. It is a figure which shows an example of a topic ratio table. It is a figure which shows an example of an item selection rate table. It is a figure which shows an example of a candidate item table. It is a figure which shows an example of a recommendation list table. It is a flowchart of the process at the time of purchase. It is a flowchart of the process at the time of item selection rate calculation. It is a flowchart of the process of step 230 of FIG. It is a flowchart of the process of step 232 of FIG. It is a flowchart of the process at the time of candidate item determination. It is a flowchart of the process at the time of recommendation list creation. It is a flowchart of a recommendation process.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 shows a configuration of an item recommendation system 1 according to an embodiment of the present invention. The item recommendation system 1 illustrated in FIG. 1 includes a purchase information processing unit 10, an item selection rate calculation unit 20, a candidate item determination unit 30, a recommendation list creation unit 40, a recommendation processing unit 50, a recording unit 60, a communication unit 70, and an input / output unit. 80. The communication unit 70 is connected to the network 3, and the input / output unit 80 is connected to the external device 4 such as an input device or a display device.

  The purchase information processing unit 10 acquires purchase information for each item in an E-Commerce (electronic commerce) service and stores the purchase information in the purchase information table 61. The item selection rate calculation unit 20 determines the topic (classification) of each purchase information stored in the purchase information table 61, and calculates the item selection rate of each item for each topic using the determined classification of each purchase information. And stored in the item selection rate table 63.

  The candidate item determination unit 30 extracts candidate items for each topic using the item selection rate of each item calculated for each topic, and stores the candidate items respectively extracted for all topics in the candidate item table 64. The recommendation list creation unit 40 calculates a recommendation score for each user for each candidate item stored in the candidate item table 64 using the item selection rate and the purchase information stored in the purchase information table 61, and recommends the recommendation A recommended list is generated by selecting recommended items recommended for each user based on the score, and stored in the recommended list table 65.

  When the recommendation processing unit 50 receives an item recommendation request from the external terminal 2, the recommendation processing unit 50 extracts a recommendation list corresponding to the user specified by the item recommendation request from the recommendation list table 65 and presents the recommendation list to the external terminal 2.

  The recording unit 60 stores a purchase information table 61, a topic ratio table 62, an item selection rate table 63, a candidate item table 64, and a recommendation list table 65. Each table will be described below.

<Purchase information table 61>
FIG. 2 is a diagram illustrating an example of the purchase information table 61. The purchase information table 61 stores purchase information related to purchased items (such as products and services) and purchased users in the E-Commerce service. The purchase information table 61 shown in the figure includes a user ID field and an item ID field. In the user ID field, the purchase information processing unit 10 sets a user identifier that identifies a user related to the purchase. In the item ID field, the purchase information processing unit 10 sets an item identifier for specifying which item the purchase relates to.

<Topic ratio table 62>
FIG. 3 is a diagram illustrating an example of the topic ratio table 62. The topic ratio table 62 indicates the strength of connection (relevance) between each user and each topic, and a record is generated for each user and topic. The topic (classification) is a classification result obtained by grouping purchase records in the purchase information table 61 into a predetermined number.

  The topic ratio table 62 shown in the figure includes a user ID field, a topic ID field, and a topic ratio field. In the topic ratio field, the probability P (z | u) that the corresponding topic z is selected in the purchase of the user u is set by the item selection rate calculation unit 20.

<Item selection rate table 63>
FIG. 4 is a diagram illustrating an example of the item selection rate table 63. The item selection rate table 63 shown in the figure includes a topic ID field, an item ID field, and an item selection rate field. In the item selection rate field, when the topic z is selected, the probability P (i | z) that the item i is purchased is set by the item selection rate calculation unit 20.

<Candidate item table 64>
FIG. 5 is a diagram illustrating an example of the candidate item table 64. The candidate item table 64 shown in the figure includes an item ID field. The item ID set in the item ID field of the candidate item table 64 is a set of items that are candidates for recommendation at the time of creating the recommendation list (a set of items for which a recommendation score is to be calculated), and is set by the candidate item determination unit 30. Is done.

<Recommendation list table 65>
FIG. 6 is a diagram illustrating an example of the recommendation list table 65. The recommendation list table 65 is a table in which items recommended for each user are set. The recommendation list table 65 shown in the figure includes a user ID field, an item ID field, and a recommendation score field. In the recommendation score field, the recommendation score of the item for the user calculated by the recommendation list creation unit 40 is set.

  As the item recommendation system 1 described above, for example, a general-purpose computer system including a CPU, a memory, and an external storage device such as an HDD can be used. In this computer system, each function of the item recommendation system 1 is realized by the CPU executing a program for the item recommendation system 1 loaded on the memory. Further, the program for the item recommendation system 1 can be stored in a computer-readable recording medium such as a hard disk, a flexible disk, a CD-ROM, an MO, a DVD-ROM, or can be distributed via a network.

  Next, specific processing of this embodiment will be described.

  FIG. 7 is a flowchart of processing when purchasing information is updated in the present embodiment. Note that the purchase information update timing may be manually managed / instructed by a system administrator, for example, or may be automatically updated when a user purchases some item.

  Step 110) The purchase information processing unit 10 acquires purchase information (purchase history information) from an E-Commerce processing unit (not shown) of the item recommendation system 1 or an E-Commerce system not shown. The E-Commerce processing unit or the E-Commerce system provides a service for selling items such as products and services to the external terminal 2 via the network 3.

  Note that the purchase information processing unit 10 may acquire purchase information from the E-Commerce processing unit or the like at the timing when an instruction from the system administrator is received, or the E-Commerce processing unit at the timing when the item is purchased. Or the like may transmit the purchase information to the purchase information processing unit 10.

  Step 120) The purchase information processing unit 10 registers the acquired purchase information in the purchase information table 61. That is, for each purchased item, a record in which the user ID of the purchased user and the item ID of the purchased item are set in the item ID field and the user ID field is registered in the purchase information table 61.

  FIG. 8 is a flowchart of processing when calculating the item selection rate in the present embodiment. The timing for calculating the item selection rate may be manually managed / instructed by a system administrator, for example, or may be automatically performed periodically.

  Step 210) The item selection rate calculation unit 20 calculates the number K of topics as parameters for recommendation, the prior parameter α for calculating the topic ratio P (z | u), and the item selection rate P (i | z). Set the prior parameter η. For these parameters, predetermined predetermined values (constants) are used. The number of topics Z designates how many topics (classifications) each purchase record in the purchase information table 61 is grouped into.

Step 220) The item selection rate calculation unit 20 refers to the purchase information table 61, and acquires purchase information, that is, user u = {u _m } _m and item i = {i _m } _{m of} each purchase record.

Step 230) The item selection rate calculation unit 20 determines the topic z = {z _m } _m of each purchase record. The process of step 230 will be described later with reference to FIG.

Step 240) The item selection rate calculation unit 20 uses the topic number K and α of the parameter set in Step 210, the topic z of each purchase record determined in Step 230, and the topic ratio P (z | u) is calculated. The topic ratio P (z | u) indicates the probability that the topic z is selected by the purchase of the user u.

  Here, m is a variable that identifies each purchase record in the purchase information table 61. The item selection rate calculation unit 20 generates a record in which the topic ratio P (z | u) is set for each user u and topic z, and registers the generated record in the topic ratio table 62. Since each purchase record includes a user ID and an item ID, a topic that is strongly associated with each user can be known.

Step 250) The item selection rate calculation unit 20 uses the parameter η set in Step 210, the topic z of each purchase record determined in Step 230, and the equation i when the topic z is selected. The item selection rate P (i | z) that is the probability of purchasing is calculated.

  Here, N is the number of unique items. The item selection rate calculation unit 20 generates a record in which the topic z, item i, and item selection rate P (i | z) are set, and registers it in the item selection rate table 63. As a result, items that are strongly associated with a certain topic can be identified.

  Next, the processing of step 230 in FIG. 8 will be described in detail.

  FIG. 9 is a flowchart of the process in step 230 of FIG.

Step 231) The item selection rate calculation unit 20 initializes the topic z _m (m = 1,..., M) of each purchase record with a random number having a minimum value 1 and a maximum value K.

  Step 232 is repeated a predetermined number of times. The number of repetitions may be an arbitrary number set in advance, or may be terminated when the topic z of each purchase record converges. FIG. 9 shows an example in which a predetermined number of times (for example, 500 times) is repeated. After the number j is initialized to 0 and S232 is performed, the number j is incremented. .

Step 232) The item selection rate calculation unit 20 updates the topic z _m (m = 1,..., M) of each purchase record. The processing in step 232 will be described in detail with reference to FIG.

  FIG. 10 is a flowchart of the process in step 232 of FIG.

  The item selection rate calculation unit 20 performs Step 2321 and Step 2322 for all the purchase records m = 1,. That is, first, m is set to 1 and steps 2321 and 2322 are performed. Then, m is incremented, and steps 2321 and 2322 are repeated until m becomes M.

Step 2321) The item selection rate calculation unit 20 calculates the sampling probability assigned to each topic candidate k = 1,..., K of the topic z _m of the purchase record using Equation 3.

  That is, first set k to 1, and after performing step 2321, increment k and repeat step 2321 until k becomes K, thereby calculating the sampling probability of each topic candidate from 1 to K. Thereafter, step 2322 is performed.

Step 2322) The item selection rate calculation unit 20 generates a random number based on the sampling probability assigned to each topic candidate k = 1,..., K of the topic z _m of the purchase record calculated in Step 2321. Update z _m . Specifically, the sampling probability is updated according to Equation 4 for k = 2,.

Further, for k = 1,..., K, the sampling is updated as shown in Equation 5.

Thereafter, a decimal random number having a minimum value of 0 and a maximum value of 1 is generated, and that value is assigned to the minimum k having a higher sampling probability. A specific example will be described. Now, assuming that K = 3, the sampling probabilities of the topic candidates calculated based on Expression 3 are P (z _m = 1) = 1.6, P (z _m = 2) = 0.4, P Assume that (z _m = 3) = 2.0. In other words, we want to sample each topic at a 4: 1: 5 ratio. At this time, using Equations 4 and 5, P (z _m = 1) = 0.4, P (z _m = 2) = 0.5, P (z _m = 3) = 1.0 and the sampling probability are updated, and the minimum value A decimal random number of 0 and maximum value 1 is generated, and that value is assigned to the minimum k that has a higher sampling probability. That is, the minimum k that has a sampling probability that exceeds the value of the generated random number is determined as the topic z _m of the purchase record.

  FIG. 11 is a flowchart of processing when a candidate item is determined. The candidate item determination timing may be instructed and managed manually by a system administrator, for example, or may be automatically performed immediately before the recommendation list is created.

  Step 310) The candidate item determination unit 30 initializes the candidate item set as C ← {}.

  And the candidate item determination part 30 performs step 320, 330, 340 about all the topics k = 1, ..., K. That is, first, k is set to 1 and steps 320, 330, and 340 are performed. Then, k is incremented, and steps 320, 330, and 340 are repeatedly performed until k becomes K.

  Step 320) The candidate item determination unit 30 refers to the record in which the value of the topic ID field in the item selection rate table 63 is k, and the item selection rate P (i of topic k for each item i = 1,. | Get z). Here, I indicates the number of unique items.

Step 330) The candidate item determination unit 30 sorts the items i = 1,..., I in descending order of the item selection rate P (i | z) for the topic _k, and acquires the top N items C _k .

Then, step 340 is performed for each item i included in C _k . That is, it is determined whether or not there is an unprocessed item i in each item i included in C _k. If there is an unprocessed item i, the unprocessed item i is selected, and the selected item i is selected. The process of step 340 is performed.

Step 340) The candidate item determination unit 30 adds each item i included in C _k to the candidate item set C (ie, C ← C∪ (i)).

  Then, after performing steps 320, 330, and 340 for all the topics k = 1,..., K, the candidate item determination unit 30 performs step 350 for each item i included in the candidate item set C. . That is, it is determined whether or not there is an unprocessed item i in each item i included in the set C. If there is an unprocessed item i, the unprocessed item i is selected, and the selected item i is selected. Step 350 is processed.

  Step 350) The candidate item determination unit 30 registers a record in which each item i included in the set C is set as the value of the item ID field in the candidate item table 64.

  In the present embodiment, by performing the processing of FIG. 11 as described above, the item selection rate calculated for each topic is used to cut off items that may not appear in any user's recommendation list, and candidate items Narrow down the number of. Thereby, the number of items for calculating the recommendation score described below can be reduced, and the cost of the recommendation list creation process can be reduced.

  FIG. 12 is a flowchart of processing when creating a recommendation list in the present embodiment. The recommendation list creation timing may be manually managed and instructed by a system administrator, for example, or may be automatically performed periodically.

  Step 410) The recommendation list creation unit 40 refers to the candidate item table 64 and acquires a candidate item set C.

  Then, the recommendation list creation unit 40 performs Step 420 for all the acquired candidate items iεC. That is, it is determined whether or not there is an unprocessed item in the candidate item set C. If there is an unprocessed item, an unprocessed item i is selected, and the process of step 420 is performed on the selected item i.

  Step 420) The recommendation list creation unit 40 refers to the item selection rate table 63 and acquires the item selection rate P (i | z) (z = 1,..., K) for each item i. Then, the recommendation list creation unit 40 performs step 430 after obtaining the item selection rate P (i | z) for all items in the candidate item set C.

  Step 430) The recommendation list creation unit 40 refers to the purchase information table 61 and obtains all user sets U.

  Then, the recommendation list creation unit 40 performs Steps 440, 450, and 460 for all the users uεU. That is, it is determined whether or not there is an unprocessed user u in the entire user set U. If there is an unprocessed user, the unprocessed user u is selected, and the process of step 440 is performed on the selected user u. Do.

  Step 440) The recommendation list creation unit 40 refers to the topic ratio table 62 and acquires the topic ratio P (z | u) (z = 1,..., K) related to the user u.

  Then, the recommendation list creation unit 40 performs Step 450 for all candidate items iεC of the candidate item set C acquired in Step 410. That is, it is determined whether or not there is an unprocessed item in the candidate item set C. If there is an unprocessed item, the unprocessed item i is selected, and the process of step 450 is performed on the selected item i.

Step 450) The recommendation list creation unit 40 uses the item selection rate acquired in Step 420 and the topic ratio acquired in Step 440 to calculate the recommendation score P (i | u) for the user u of the item i according to Equation 6. calculate.

  Then, after calculating the recommendation score P (i | u) for all items in the candidate item set C, the recommendation list creation unit 40 performs Step 460.

  Step 460) Based on the recommendation score P (i | u) (i∈C) of each item calculated in step 450, the recommendation list creation unit 40 sets any item as a recommended item for the user u. To decide. Then, the recommendation list creation unit 40 generates a record in which u is set in the user ID field, i is set in the item ID field, and P (i | u) is set in the recommendation score field for each determined recommended item. Register with. Note that since the user ID and the recommended item are stored in association with each other in the record of the recommendation list table 65, the recommendation list is generated by extracting the record with the user ID.

  As a criterion for determining whether or not to be a recommended item, for example, an arbitrary number of items set in advance in descending order of recommended score values are determined as recommended items, or some threshold is set for the recommended score. For example, it may be possible to determine an item whose recommended score exceeds the threshold as a recommended item.

  Steps 440, 450, and 460 are repeated until there are no unprocessed users. By the processing of FIG. 12 described above, items (recommendation list) recommended for each user are set in the recommendation list table 65.

  FIG. 13 is a flowchart of processing during recommendation processing in the present embodiment. The recommendation process is executed, for example, when an item recommendation request is transmitted from the external terminal 2 to the item recommendation system 1 through the communication unit 70 and the network 3.

  Step 510) The recommendation processing unit 50 receives an item recommendation request regarding the user u from the external terminal 2. This request includes the user ID of the target user u as an argument.

  Step 520) The recommendation processing unit 50 extracts all recommended items i corresponding to the user ID specified in the request from the recommendation list table 65, generates a recommendation list for the user u, By presenting the recommendation list to the terminal 2, the item is recommended to the user u.

  In the present embodiment described above, in order to recommend an item with an Internet service with a very large number of items, which user's recommendation list is used by using an intermediate result of recommendation processing (such as an item selection rate for each topic). By cutting off items that do not appear in the list and narrowing down candidate items, only the recommendation score for the candidate item is calculated, and a recommendation list is created at high speed. That is, a recommendation list can be created at high speed by reducing the number of combinations of items for calculating recommendation scores and users.

  As a result, in this embodiment, while maintaining the top items selected as recommended items for each user, the items can be recommended at high speed in services on the Internet with a large number of items and users. Become. That is, the calculation cost and processing load in item recommendation can be reduced. This facilitates the purchase of large-scale services on the Internet, leading to a significant increase in revenue.

  In addition, this invention is not limited to the said embodiment, Many deformation | transformation are possible within the range of the summary.

1: Item recommendation system 10: Purchase information processing unit 20: Item selection rate calculation unit 30: Candidate item determination unit 40: Recommendation list creation unit 50: Recommendation processing unit 60: Storage unit 61: Purchase information table 62: Topic ratio table 63 : Item selection rate table 64: Candidate item table 65: Recommendation list user table 70: Communication unit 80: Input / output unit 2: External terminal 3: Network 4: External device

Claims (5)

An item recommendation system,
Purchase information acquisition means for acquiring purchase information for each item and storing it in the purchase information storage means;
The classification of each purchase information stored in the purchase information storage means is determined, the item selection rate of each item is calculated for each classification using the determined classification of the purchase information, and stored in the item selection rate storage means Item selection rate calculation means,
Using the item selection rate of each item calculated for each category, for each category, an item with a high item selection rate among the items belonging to each category is extracted as a candidate item, and extracted for each category. Candidate item determining means for storing candidate items in the candidate item storage means;
For each candidate item stored in the candidate item storage means, a recommendation score is calculated for each user using the item selection rate and the purchase information stored in the purchase information storage unit, and the user is based on the recommendation score. An item recommendation system comprising: a recommendation list creation unit that selects a recommended item for each recommendation, generates a recommendation list, and stores the recommendation list in a recommendation list storage unit. The item recommendation system according to claim 1,
When the item recommendation request is received from the terminal, the recommendation list corresponding to the user specified in the item recommendation request is extracted from the recommendation list storage unit, and is further provided with a recommendation processing unit that is presented to the terminal. Item recommendation system. An item recommendation method performed by a computer,
A purchase information acquisition step of acquiring purchase information for each item and storing it in a purchase information storage unit;
The classification of each purchase information stored in the purchase information storage unit is determined, the item selection rate of each item is calculated for each category using the determined classification of the purchase information, and stored in the item selection rate storage unit An item selection rate calculation step;
Using the item selection rate of each item calculated for each category, for each category, an item with a high item selection rate among the items belonging to each category is extracted as a candidate item, and extracted for each category. A candidate item determination step of storing the candidate item in the candidate item storage unit;
For each candidate item stored in the candidate item storage unit, a recommendation score is calculated for each user using the item selection rate and the purchase information stored in the purchase information storage unit, and the user is based on the recommendation score. An item recommendation method comprising: selecting a recommended item for each recommendation, generating a recommendation list, and storing the recommendation list in a recommendation list storage unit. The item recommendation method according to claim 3,
When an item recommendation request is received from a terminal, a recommendation list corresponding to the user specified in the item recommendation request is extracted from the recommendation list storage unit, and a recommendation processing step of presenting to the terminal is further performed. Item recommendation method. An item recommendation program for causing a computer to function as the item recommendation system according to claim 1.

Images