JP5309182B2 - Item recommendation method and item recommendation program - Google Patents

Description

  The present invention relates to a technique for selecting an item recommended for a user.

  In many electronic commerce sites, collaborative filtering-based recommendation engines have been introduced, which are effective in sales of products. On the other hand, since the current recommendation engine performs a process of predicting the current potential interest based on items purchased by the user in the past, only items similar to items purchased by the user in the past are often recommended. . For example, to a user who has purchased a certain metal lock item A, the current recommendation engine recommends the same metal lock item or another item of the same artist. However, it has been pointed out that such a recommendation is easily bored for the user, and the satisfaction level of the user decreases after the initial stage of using the recommendation. Therefore, Non-Patent Document 1 has developed a technique for diversifying recommended items based on the distribution of items on the taxonomy.

Cai-Nicolas Ziegler, Sean M. McNee, Joseph A. Konstan, and Georg Lausen, "Improving Recommendation Lists Through Topic Diversification", Proc. WWW'05, 2005, p.22-32

  However, the current technology for diversifying recommended items based on taxonomies checks whether the items are widely distributed on the taxonomy or not by checking the distribution by sequentially switching items in the recommendation list and items outside the recommendation list. Since the combination brute force process is performed to select the item with the largest value, the calculation cost is increased for diversifying the items in the recommendation list, which impedes the practical application.

  The present invention has been made in view of the above, and an object thereof is to diversify items in a recommendation list presented to a user.

In the item recommendation method according to the first aspect of the present invention, in the hierarchical structure of categories to which items belong, the total distance between categories to which each item belongs in a list composed of a plurality of items is defined as diversity of the list. An item recommendation method executed by a computer that outputs the list that maximizes the diversity, the step of inputting a recommendation list composed of a plurality of items and an out-of-service list, and the stored in the storage means Calculating a diversity of the recommendation list with reference to a hierarchical structure, calculating a diversity of the recommendation list when an item in the recommendation list and an item in the out-of-service list are exchanged, and the recommendation list Compares the diversity of the recommendation list with the diversity of items and the combination that maximizes the diversity And having the steps of replacing Align with an item of the recommendation list items within said service area list, and outputting the recommendation list interchanged, the.

  In the item recommendation method, the calculating step selects a first item to be put into the recommendation list from the out-of-service list, and totals the distances from the category of the first item to the category of each item in the recommendation list. Calculating the first hop number, and for each item in the recommendation list, the category of each item in the recommendation list when the first item is included in the recommendation list from the category of the item Calculating a second number of hops that is the sum of distances to, selecting a second item that minimizes the second number of hops, and calculating a first diversity of the recommendation list And a step of calculating a second diversity when the first item is put in the recommendation list and the second item is removed. And the step of exchanging replaces the second item and the first item in the recommendation list when the second diversity is greater than the first diversity. It is characterized by.

  In the item recommendation method, the step of selecting the second item includes selecting the second item for each category number to which each item in the recommendation list belongs, and calculating the second diversity. , Calculating the second diversity for each of the second items selected for each category number, and the step of replacing is such that the largest of the second diversity is greater than the first diversity. Is larger, the second item giving the maximum second diversity and the first item are exchanged.

  An item recommendation program according to a second aspect of the present invention causes a computer to execute the item recommendation method.

  According to the present invention, the items in the recommendation list presented to the user can be diversified.

It is a figure which shows the example of a taxonomy. It is a functional block diagram which shows the structure of the item recommendation apparatus in this Embodiment. It is a figure which shows the example of the taxonomy information accumulate | stored in the said item recommendation apparatus. It is a figure which shows the example of the recommended item information input into the said item recommendation apparatus. It is a flowchart which shows the flow of a process of the item recommendation apparatus in this Embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

In the present embodiment, for items extracted based on the user's interest, items in the upper list L and in the out-of-service list M so that the diversity of the list of items recommended to the user (upper list L) is maximized. Swap items. Diversity (hereinafter referred to as “Diversity”) is a value indicating how widely the instances (items) in the recommendation list are distributed on the ontology. In the present embodiment, the diversity calculation formula is defined as the following formula (1).

  Here, H (i) is the number of hops to other items in the upper list L viewed from the item i, and | C (i) | is the number of classes to which the item i belongs. The number of hops is the shortest distance on the taxonomy (class structure).

FIG. 1 shows an example of a taxonomy. In the figure, a taxonomy having a hierarchical structure having _seven classes (categories) C _{1 to} C ₇ and six instances (items) Ia to Ic and Ix to Iz are shown. A class is connected to other classes in a parent-child relationship to form a hierarchical structure, and an instance is a node whose parent is a class. For example, class C ₂ has class C ₁ as a parent class and classes C ₄ and C ₅ as child classes. Item Ia is an instance, the class C ₄ parent class. There is also an item that has a plurality of classes C ₅ and C ₆ as a parent class like the item Ib, and the class number | C (Ib) | of the item Ib at this time is 2.

The number of hops can be obtained by counting the classes that are traced when the class is traced from one item to the other item in the shortest time. For example, since the item Ia can be traced to the class I ₄ -C ₂ -C ₁ -C ₃ -C ₇ , the number of hops is 4. In this case, the parent class C ₄ items Ia is not included in the number of hops.

  Diversity when the items Ia, Ib, and Ic are included in the upper list L is calculated by the following equation and becomes 21.

{(2 + 4 + 3) / 1} + {(2 + 3) / 2} + {(4 + 1) / 2} + {(3 + 1 + 3) / 1} = 21
For example, the first term on the left side is obtained by dividing the total number of hops from the item Ia to the items Ib and Ic by the number of classes to which the item Ia belongs. Since the item Ib belongs to a plurality of classes C ₅ and C ₆ , it is calculated as a separate item. Therefore, the first term on the left-hand side includes the number of hops 2 that reach the item Ib by following the class C ₄ -C ₂ -C ₅ from the item Ia, and the class C ₄ -C ₂ -C ₁ -C _3- from the item Ia. Item Ia is obtained by summing up the number of hops 4 that follow C ₆ and reach item Ib, and the number of hops ₃ that follow class C ₄ -C ₂ -C ₁ -C ₃ and reach item I _c from item Ia. Divided by the number of classes that belong to. The second term on the left side is the sum of the number of hops to the items Ia and Ic seen from the item Ib belonging to the class C ₅ divided by the number of classes 2 to which the item Ib belongs, and the third term on the left side Is the sum of the number of hops from the item Ib belonging to the class C ₆ to the items Ia and Ic divided by the number of classes 2 to which the item Ib belongs. The fourth term on the left side is the total number of hops from the item Ic to the items Ia and Ic.

  Next, the item recommendation device in the present embodiment will be described.

  FIG. 2 is a functional block diagram showing the configuration of the item recommendation device in the present embodiment. The item recommendation device 1 shown in FIG. 1 includes an input unit 11, a calculation unit 12, a replacement unit 13, a taxonomy storage unit 14, and a recommended item storage unit 15. Each unit included in the item recommendation device 1 may be configured by a computer including an arithmetic processing device, a storage device, and the like, and the processing of each unit may be executed by a program. This program is stored in a storage device included in the item recommendation device 1, and can be recorded on a recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or provided through a network.

  The input unit 11 inputs recommended item information, divides the input recommended item information into an upper list L and an out-of-service list M, and causes the recommended item storage unit 15 to store the recommended item information.

  The calculation unit 12 calculates Diversity when the items in the upper list L and the items in the upper list L and the items in the out-of-service list M are exchanged. At the time of Diversity calculation, a combination of replacements that is expected to have the same or smaller Diversity is skipped. Details of Diversity calculation will be described later.

  The replacement unit 13 switches the items in the upper list L and the items in the out-of-service list M so that the Diversity calculated by the calculation unit 12 is maximized.

  The taxonomy accumulating unit 14 accumulates taxonomy information. FIG. 3 shows an example of taxonomy information. The taxonomy information includes information on classes constituting the taxonomy. The class information includes a class ID for identifying a class, child classes, and parent class information.

  The recommended item storage unit 15 stores recommended item information. FIG. 4 shows an example of recommended item information. The recommended item information includes item data, and the item data includes an item ID for identifying the item and parent class information. The recommended item storage unit 15 stores the item data separately in the upper list L and the out-of-service list M.

  Next, the flow of processing of the item recommendation device of this embodiment will be described. Here, the example shown in FIG. 1, that is, the example in which the taxonomy information shown in FIG. 3 is stored in the taxonomy storage unit 14 and the recommended item information shown in FIG. 4 is input will be described.

  FIG. 5 is a flowchart showing a process flow of the item recommendation device 1 according to the present embodiment.

  First, the input unit 11 inputs recommended item information (step S101), and the input recommended item information is divided into an upper list L and an out-of-service list M and stored in the recommended item storage unit 15 (step S102). Specifically, the number of items in the list and the number of out-of-service list items are obtained from the initial setting file stored in the memory or the like in the item recommendation device 1, and the recommended item information is the number of items in the list and the number of out-of-service list items Are entered in descending order of recommendation to the user. Then, with the number of items in the list as a threshold, the recommended item information is stored in the recommended item storage unit 15 as the upper list L from the first to the threshold of the acquired recommended item information and as the out-of-service list M after the threshold +1. For example, if the number of items in the list and the number of out-of-service list items are both 3, that is, if the number of items in the upper list L and the number of items in the out-of-service list M are both, 6 items Ia to Ic and Ix to Iz are input. The items Ia to Ic are stored in the recommended item storage unit 15 as the upper list L and the items Ix to Iz are stored as the out-of-service list M. It should be noted that the recommended item information is extracted by arranging the items in the order in which the user is likely to be interested based on the user's interest by another recommended item extraction process.

  When the input of the recommended item information is completed, the following processing is repeated in order from the first in the out-of-service list M for all items in the out-of-service list M. Hereinafter, the ii-th (initial value ii = 1) item M (ii) will be described as an item in the out-of-service list M to be processed.

  First, the calculation unit 12 refers to the taxonomy storage unit 14 and adds the ii-th item M (ii) of the out-of-service list M to the upper list L, and the hop count H (L) and class count C {M (Ii)} is calculated (steps S103 and S104). A list obtained by adding the ii-th item of the out-of-service list M to the upper list L is represented as {L + M (ii)}. The number of hops H (L) is the total number of hops with other items in the upper list L as seen from the item M (ii). The class number C {M (ii)} is the number of classes to which the item M (ii) belongs. For example, when the items Ia to Ic exist in the upper list L and the item Ix of the out-of-service list M is added to the upper list L, the number of hops from the item Ix to the item Ia is 4, and the number of hops from the item Ix to the item Ib 4 and 2, the number of hops from item Ix to item Ic is 1, and the total number of hops is 11. The number of classes of item Ix is one.

  Subsequently, the calculation unit 12 sets the item to be removed as L (i) and sequentially removes the i-th item L (i) in the upper list L from the list {L + M (ii)} in order from i = 1. The total number of hops H {L + M (ii) −L (i)} to each item in the list {L + M (ii) −L (i)} viewed from the item L (i), the number of classes C {L (i) } Is calculated (steps S105 and S106). This process is performed for all items in the upper list L (step S107).

For example, when the items Ia to Ic exist in the upper list L and the item Ix of the out-of-service list M is added to the upper list L, the items L ( The total number of hops to each item in the list {L + M (ii) −L (i)} viewed from i) and the number of classes of L (i) are as shown in Table 1 below.

  Then, the item with the smallest total number of hops is picked up for each class number of items in the upper list L (step S108). For example, in the case of Table 1 above, the item with the class number of 1 and the smallest hop number is the item Ic, and the item with the class number of 2 is only the item Ib, so the items Ib and Ic are picked up. If the number of classes and the number of hops are the same, the one with the higher rank, that is, the one recommended to the user is picked up.

  Here, for all picked up items, H (L) = <H {L + M (ii) −L (i)} and C {M (ii)}> = C {L (i)} Advances ii by one and returns to step S103 (step S109). When the item L (i) satisfying the above condition is replaced with the item (ii) in the out-of-service list M, it is clear that Diversity is reduced. Therefore, if all the items that have been picked up satisfy the above condition, ii is incremented by one without processing Diversity, and the item M (ii) in the next out-of-service list M is processed.

  Then, the calculation unit 12 obtains Diversity {L + M (ii) −L (i)} in the case where the diversity {L} of the upper list L, the item M (ii) and the item L (i) picked up in step S108 are replaced. Is calculated (step S110). Note that the item L (i) that satisfies the condition of step S109 is excluded from Diversity {L + M (ii) −L (i)} calculation.

For example, Diversity {L} when item Ia, item Ib, and item Ic are included in the upper list L is 21. When the item Ix and the item Ib are interchanged, and when the item Ix and the item Ic are interchanged, the respective Diversity is as shown in Table 2 below.

  Here, when Diversity {L} is larger than the maximum of the calculated Diversity {L + M (ii) −L (i)}, ii is advanced by 1 and the process returns to step S103, and the next item M Processing is performed for (ii) (step S111).

  Then, the replacement unit 13 replaces the item L (i) that maximizes the calculated Diversity {L + M (ii) −L (i)} with the item M (ii) (step S112). In Table 2 above, Diversity when the item Ic is replaced is the maximum and is greater than Diversity {L} = 21. Therefore, the item Ic is removed from the upper list L, and the item Ix is inserted. Update information. Since items removed from the upper list L are not recommended, they are not returned to either the upper list L or the out-of-service list M. At this point, the upper list L includes items Ia, Ib, and Ix, and the out-of-service list M includes items Iy and Iz.

  If an item still remains in the out-of-service list M, ii is advanced by 1 and the process returns to step S103 to process the next item M (ii) (step S113). When all items in the out-of-service list M are processed, the items in the upper list L are output as recommended items recommended to the user.

  As described above, according to the present embodiment, the Diversity calculation formula of a list composed of a plurality of items represents the total distance on the taxonomy from the class to which the item belongs to the class to which another item belongs. It is possible to present a variety of recommended items to the user by changing the items in the upper list L and the items in the out-of-service list M so as to maximize the Diversity by defining as in the formula (1) It becomes. As a result, the user who uses the service can be given an opportunity to know various items presented by the service, and the user can be prevented from getting bored with the service. Moreover, since various items can be recommended, the user's willingness to purchase can be increased and the sales can be increased.

  According to the present embodiment, in the calculation of Diversity when the item M (ii) in the out-of-service list M is replaced with the item L (i) in the higher-level list L, the item L (i) in the higher-level list L When the item M (ii) in M is replaced, the total number of hops to each item in the list {L + M (ii) −L (i)} viewed from the item L (i) is calculated. For each class number of (i), Diversity is calculated only for the smallest total number of hops, and is compared with the largest Diversity calculated among the Diversity of the upper list L and replaced with item M (ii). By determining whether or not, the diversity is maximized without calculating the diversity for all the combinations. Replacement of items is possible.

DESCRIPTION OF SYMBOLS 1 ... Item recommendation apparatus 11 ... Input part 12 ... Calculation part 13 ... Replacement part 14 ... Taxonomy accumulation | storage part 15 ... Recommended item accumulation | storage part

Claims (4)

In the hierarchical structure of the category to which the item belongs, the total distance between the categories to which each item in the list consisting of multiple items belongs is defined as the diversity of the list, and the list that maximizes the diversity is output. An item recommendation method executed by a computer ,
Inputting a recommendation list composed of a plurality of items and an out-of-service list;
Referencing the hierarchical structure stored in the storage means, calculates the diversity of the recommendation list, and calculates the diversity of the recommendation list when the items in the recommendation list and the items in the out-of-service list are exchanged And steps to
Comparing the diversity of the recommendation list with the diversity of the recommendation list when the item is replaced, and replacing the item in the recommendation list and the item in the out-of-service list with a combination that maximizes the diversity;
Outputting the replaced recommendation list;
An item recommendation method characterized by comprising: The calculating step includes:
Selecting a first item to be put into the recommendation list from the out-of-service list, and calculating a first hop number that is a total distance from the category of the first item to the category of each item in the recommendation list; ,
For each item in the recommendation list, a second hop number that is the total distance from the category of the item to the category of each item in the recommendation list when the first item is put in the recommendation list. A calculating step;
Selecting a second item that minimizes the second hop number;
Calculating a first diversity of the recommendation list, and calculating a second diversity when the first item is put in the recommendation list and the second item is removed. ,
The step of replacing includes replacing the second item and the first item in the recommendation list when the second diversity is greater than the first diversity. Item recommendation method according to 1. The step of selecting the second item selects the second item for each category number to which each item in the recommendation list belongs,
The step of calculating the second diversity calculates the second diversity for each case of the second item selected for each category number;
The step of exchanging includes the second item that gives the second diversity that is the maximum when the maximum of the second diversity is larger than the first diversity, and the first item The item recommendation method according to claim 2, wherein the item is replaced.   An item recommendation program for causing a computer to execute the item recommendation method according to any one of claims 1 to 3.

Images