JP2012079268A - Recommended item retrieval server and recommended item retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable identification of recommended items in real time without requiring high capacity memory.SOLUTION: A recommended item retrieval server 10 for identifying recommended items to be recommended to a user visiting a purchase site to purchase them based on purchase history information showing item purchase history of each user comprises a purchase history information database 10A in which the purchase history information is stored. The server identifies related users who are other users each having a purchase history of the visiting user related item showing an item which the visiting user purchased or will purchase by referring to the purchase history information, calculates an item set which is a set of items purchased by the identified related users by referring to the purchase history information, derives a correlation degree between the visiting user related item and each item belonging to the derived item set, and identifies a prescribed number of items each of whose correlation degree with the visiting user related item is high as recommended items according to the derived correlation degree.

Description

  The present invention relates to a recommended item search server and a recommended item search program for specifying recommended items recommended to purchase for visiting users who have visited a purchase site based on purchase history information indicating item purchase history of each user.

  2. Description of the Related Art Conventionally, various Internet shop sites that provide Internet shops, which are virtual shops that can purchase products on the Internet, have been operated.

  Some Internet shop sites have a function of identifying a product recommended for purchase by a visiting user who has visited the site and presenting the recommended product. For example, based on the purchase history of the visiting user and the purchase history of each user, a product recommended for purchase by the visiting user is specified.

  A filtering technique called item-based collaborative filtering (or user-based collaborative filtering) is used as a method for specifying products recommended for purchase (see, for example, Non-Patent Document 1).

  In this item-based collaborative filtering, based on the purchase history information for all items, a value indicating the correlation between the items is calculated and stored in advance, and recommended products are specified using the stored data.

“Amason.com Recommendations”, [online], [searched September 30, 2010], Internet <URL: http://hugo.csie.ntu.edu.tw/~yjhsu/courses/u2010/papers/Amazon % 20Recommendations.pdf # search = 'Amazon% 20Recommendation% 202003% 20IEEE'>

  However, in the conventional filtering technique described above, it is necessary to calculate in advance a value indicating the correlation between each item. However, this calculation process takes an enormous amount of time (for example, several hundreds for 100,000 items). Therefore, it was necessary to calculate in advance before performing the recommended item specifying process. Therefore, there is a problem that the recommended item cannot be specified based on the latest history information and the real-time property is lacking.

  In addition, although the value indicating the correlation between items has a huge amount of data, there is a problem that a large-scale memory is required because the data needs to be stored continuously.

  An object of the present invention is to solve the above-described problems and to specify a recommended item in real time without requiring a large-capacity memory.

  The recommended item search server of the present invention is a recommended item search server that identifies recommended items recommended for purchase to visiting users who have visited a purchase site based on purchase history information indicating an item purchase history of each user. , Purchase history information storage means for storing the purchase history information, and purchase of a visitor-related item that refers to the purchase history information and indicates an item purchased by the visiting user or an item that the visiting user plans to purchase A related user specifying means for specifying a related user who is another user having a history, and an item set which is a set of items purchased by the related user specified by the related user specifying means with reference to the purchase history information The item set deriving means, the visiting user relation item, and the item set deriving means. Correlation degree deriving means for deriving the degree of correlation with each item belonging to the item set derived in the above, and the degree of correlation with the visiting user relation item according to the degree of correlation derived by the correlation degree deriving means And a recommended item specifying means for specifying a high predetermined number of items as the recommended items.

  With the above configuration, the recommended item can be specified in real time without requiring a large-capacity memory.

  The purchase history information is stored in a purchase history matrix table having an item axis in which each item is arranged and a user axis in which each user is arranged, and whether or not there is a purchase record in each square in the purchase history matrix table May be configured to store purchase record information indicating.

  The related user specifying means confirms each square corresponding to the visiting user related item in the item axis of the purchase history matrix table, and stores purchase result information indicating that there is a purchase record of the visited user related item. The user on the user axis corresponding to each square may be configured as a related user.

  The item set deriving means may be configured to check each cell corresponding to the related user in the user axis of the purchase history matrix table and derive an item set that is a set of items purchased by the related user. Good.

  It may be configured to include item recommendation means for recommending the recommended item specified by the recommended item specifying means to the visiting user.

  The recommended item search program according to the present invention is a recommended item search program for identifying recommended items recommended for purchase by visiting users who have visited a purchase site based on purchase history information indicating an item purchase history of each user. And referring to the purchase history information stored in the purchase history information storage means for storing the purchase history information in the computer, the item purchased by the visiting user or the item scheduled to be purchased by the visiting user. A related user specifying step for specifying a related user who is another user having a purchase history of the visiting user related item to be displayed, and an item of the item purchased by the related user specified in the related user specifying step with reference to the purchase history information An item set deriving step for deriving an item set that is a set, and the visit user Correlation degree deriving step for deriving the degree of correlation between the relationship item and each item belonging to the item set derived in the item set deriving step, and according to the degree of correlation derived in the correlation degree deriving step, The recommended item specifying step of specifying a predetermined number of items having a high degree of correlation with the visiting user-related items as the recommended items is executed.

  According to the present invention, a recommended item can be specified in real time without requiring a large-capacity memory.

It is a block diagram which shows the structural example of a recommendation item search system. It is a block part which shows the example of a structure of a recommendation item search server. It is explanatory drawing which shows the example of the storage state of purchase history information. It is explanatory drawing for demonstrating the existing method for providing a recommendation item. It is a flowchart which shows the example of a recommendation item specific process. It is explanatory drawing which shows the example of an item search screen. It is explanatory drawing for demonstrating the utilization method of the purchase history information in a recommendation item specific process. It is explanatory drawing which shows examples, such as a correlation derived | led-out by recommendation item specific processing. It is explanatory drawing which shows the example of a recommendation item display screen. It is a flowchart which shows the other example of a recommendation item specific process. It is explanatory drawing which shows the example of an item browsing screen.

  Hereinafter, an example of an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration example of a recommended item search system 100 according to an embodiment of the present invention. As shown in FIG. 1, a recommended item search system 100 includes a recommended item search server 10, user terminals 21 to 2N (N: an arbitrary positive integer), a base station 30, and a Web front end server (hereinafter referred to as WFE). Server) 40.

  The base station 30 and the WFE server 40 are connected to a communication network 50 such as a dedicated line or the Internet. Further, the user terminals 21 to 2N are connected to the base station 30 through wireless communication lines. That is, the user terminals 21 to 2N are connected to the communication network 50 via the base station 30. The recommended item search server 10 is connected to the WFE server 40 that plays a role of receiving requests from end users (in this example, user terminals 21 to 2N) through a dedicated line. Hereinafter, regarding the communication between the user terminals 21 to 2N and the recommended item search server 10, the description of the operations of the base station 30 and the WFE server 40 is omitted.

  The recommended item search server 10 is configured by a general information processing apparatus including an OS (Operating System), middleware, and various applications. The recommended item search server 10 is a server managed by a system administrator of the recommended item search system 100, for example.

  The recommended item search server 10 performs a process (recommended item identification) for recommending an item to a user based on various information including purchase history information (or product purchase history) stored in an information storage medium that the recommended item search server 10 has. A function for executing various processes including (processing). The purchase history information and the recommended item specifying process in this example will be described later in detail (see FIGS. 3 and 5).

  FIG. 2 is a block diagram showing an example of the configuration of the recommended item search server 10 in this example. As shown in FIG. 2, the recommended item search server 10 includes a control unit 11, a related user specifying unit 12, an item set deriving unit 13, a correlation degree deriving unit 14, a recommended item specifying unit 15, and an item recommending unit. 16 and the like.

  The control unit 11 is configured by, for example, a CPU (Central Processing Unit), and is a computer program (a computer program for causing the navigation device 100 to control the operation) stored in an information storage medium included in the recommended item search server 10. In accordance with a program), each element constituting the recommended item search server 10 is comprehensively controlled, and has various functions such as a function for executing an information search process for searching various information. Note that the various functions of the control unit 11 are realized by known techniques included in a general server device.

  The related user specifying unit 12 refers to a purchase history information DB (database) 10A, which will be described later, and stores a purchase history of “visiting user related items” indicating items purchased by the visiting user or items that the visiting user plans to purchase. Identify other users (related users) you have

  The item set deriving unit 13 has a function of performing processing for deriving an item set that is a set of items purchased by the related user specified by the related user specifying unit 12 with reference to a purchase history information DB 10A described later. An example of the item set derivation method will be described in detail later.

  The correlation degree deriving unit 14 has a function of performing a process of deriving a degree of correlation between the visiting user-related item and each item belonging to the item set derived by the item set deriving unit 13. Here, the “correlation degree” in this example indicates a probability that a user who has a purchase history of an item (item 1) purchases another specific item (item 2). . That is, it indicates the probability (so-called co-occurrence frequency) that two items are simultaneously included in the purchase history of the same user. As a method for deriving the degree of correlation, a known arithmetic expression such as a cosine processing measurement method that indicates the degree of correlation by a cosine distance can be used. Note that the “cosine distance” is a value obtained by dividing the inner product of two vectors by the product of the absolute values of the vectors when the feature quantities of two elements are each represented by a vector.

  In this example, the correlation degree deriving unit 14 includes an item purchased by the user (user purchase item) “t” and an item (co-occurrence item) purchased by another user who purchased the item purchased by the user. The correlation between the user purchase item “t” and the co-occurrence item “r” is derived by using Equation 1 below with respect to “r”.

  The recommended item specifying unit 15 recommends a predetermined number of items having a high degree of correlation with the visiting user relation item (for example, within 5 in descending order of the degree of correlation) according to the degree of correlation derived by the degree of correlation deriving unit 14. It has a function of performing a process specified for an item.

  The item recommendation unit 16 has a function of performing a process for recommending the recommended item specified by the recommended item specification unit 15 to the visiting user. In this example, the item recommendation unit 16 recommends a recommended item by displaying the recommended item on a display unit included in the user terminal. Specifically, the item recommendation unit 16 creates data for displaying, for example, a display screen including a recommended item display area for displaying the name of the recommended item on the display screen of the display device included in the user terminal, A function of performing processing of transmitting the created data to the user terminal;

  The recommended item search server 10 also includes a purchase history information DB 10A that is a storage medium for storing purchase history information.

  FIG. 3 is an explanatory diagram illustrating an example of a storage state of purchase history information stored in the purchase history information DB 10A. As shown in FIG. 3, the purchase history information in this example is stored in a purchase history matrix table having an item axis in which each item is arranged and a user axis in which each user is arranged. In each cell, purchase record information indicating the presence / absence of purchase record is stored. Specifically, as shown in FIG. 3, the purchase history information has an item ID that can uniquely identify an item as a column item, and a matrix format that has a user ID that can uniquely identify a user as a row item. For example, when the user indicated by the user ID “U1” purchases the item indicated by the item ID “I1”, “1” is set in the (U1, I1) element of the matrix. It is assumed that “0” is set for the other elements. Further, an element for which “1” is set may be configured such that “0” is set again when a predetermined condition (for example, a predetermined number of days have passed since the item purchase date) is satisfied. Hereinafter, in order to facilitate the description in this example, the purchase history information will be described using a “5 × 5” matrix as an example.

  Next, the operation of the recommended item search server 10 in the recommended item search system 100 will be described with reference to the drawings. Note that the contents of operations and processes not particularly related to the present invention may be omitted.

  In this example, the recommended item search server 10 recommends a purchase to a visiting user who has visited a purchase site based on purchase history information indicating an item purchase history of each user who operates the user terminals 21 to 2N. The recommended item specifying process for specifying is executed. The purchase site may be configured to be provided by the recommended item search server 10, or the recommended item search server 10 may provide information related to a user visit at the purchase site provided by the purchase site creator registered in the recommended item search server 10. It is good also as a structure which collects.

  Here, the difference between the existing method used for providing the recommended item again and the recommended item specifying process executed by the recommended item search server 10 will be described.

  FIG. 4 is an explanatory diagram for explaining an existing method for providing a recommended item. As shown in FIG. 4, in the existing method, the neighborhood information indicating the degree of correlation between the item groups included in the purchase history information is collectively calculated from the purchase history information by batch processing. For this reason, the processing burden associated with this processing is large, and depending on the size of the purchase history information, it may take several tens of hours from the start to the end of the processing. Therefore, when the recommended item is provided based on the local information and the item purchase history of the visiting user, the time lag between the user's visit (that is, when browsing the site) and the update of the neighborhood information becomes large. There was a problem. There is also a problem that it is necessary to secure a large-scale storage area for storing the neighborhood information.

  In contrast to such an existing method, the recommended item specifying process executed by the recommended item search server 10 is technically significant in that it does not require creation and storage of neighborhood information. That is, the recommended item specifying process executed by the recommended item search server 10 is useful in that a recommended item is searched by deriving a direct correlation degree from input information for each recommendation request from purchase history information.

  FIG. 5 is a flowchart illustrating an example of recommended item specifying processing executed by the recommended item search server 10. In the following description, the processing subject is described as the control unit 11 included in the recommended item search server 10.

  The recommended item specifying process is performed by displaying a screen for accepting a general genre search or a keyword search as shown in FIG. 6 on the display screen of the display device of the user terminal 22 that has accessed the recommended item search server 10. When an item search screen including a search button 101 for receiving a request and a recommended item display button 102 for receiving a display request for a recommended item display screen is displayed, the recommended item search server 10 displays a recommended item from the user terminal 22. The process is started in response to receiving a notification that the selection of the button 102 has been accepted. Hereinafter, the operation of the recommended item search server 10 for the user terminal 22 will be described.

  In the recommended item specifying process, first, the control unit 11 specifies a purchase history (information indicating an item purchased by the user) of the user of the user terminal 22 who is a visiting user, and specifies an input matrix T (user ID) ( Step S101). In this example, the case where the user ID indicating the user of the user terminal 22 is “U2” will be described as an example (see FIG. 3).

  FIG. 7 is an explanatory diagram for explaining a method of utilizing purchase history information in the recommended item specifying process. Hereinafter, description will be made with reference to FIG. 7 as appropriate. In order to facilitate the explanation below, the letter “ID” is omitted when an item or a user is indicated. That is, for example, the user indicated by the user ID “U2” is expressed as “user U2”, and the item indicated by the item ID “I1” is expressed as “item I1”. When user information or item information constitutes an element of a matrix, each element is represented by a corresponding ID, for example, “I1”.

  Here, the control part 11 specifies element [I1, I3] as an element of input matrix T (U2). Each element of the input matrix is a “visiting user-related item”.

  When the input matrix is specified (see FIG. 7A), the control unit 11 determines whether or not the element of the input matrix is empty (step S102). Here, when it is determined that the element of the input matrix is empty (Y in step S102), the control unit 11 proceeds to a process in step S110 described later.

  On the other hand, when determining that the elements of the input matrix are not empty (for example, determining that the elements of the input matrix T (U2) are [I1, I3]) (N in step S102), the control unit 11 One of them is specified as a user purchase item (that is, one of the elements of the input matrix T (U2) is substituted for “t”) (step S103). The element selection method here is not particularly limited. For example, when the item ID is indicated by a numerical value, the element having the smallest numerical value may be selected.

  For example, when the element “I1” is substituted for “t”, the control unit 11 deletes the substituted element “I1” from the input matrix T (U2) (step S104). That is, hereinafter, the element of the matrix T (U2) is [I3].

  When the substituted element is deleted from the input matrix, the control unit 11 causes the item set deriving unit 13 to specify the item set R (t) co-occurring with “t” (step S105).

  In this example, first, the control unit 11 uses the related user specifying unit 12 to specify another user RU (t) (hereinafter referred to as related user RU (t)) having a purchase history of items purchased by the user U2. (See FIG. 7B). Here, the element [U1, U2, U3, U4, U5] is specified as the related user RU (I1). Next, the control unit 11 deletes the visiting user U2 from the related user RU (I1) specified by the related user specifying unit 12, and the item set deriving unit 13 shares the item I1 with the item I1 based on the purchase history of each related user. The raising item set R (I1) is specified. Here, the element [I2, I3, I4, I5] is derived as an item set R (I1) (see FIG. 7C).

  Note that the item set deriving unit 13 may be configured to exclude an item (in this example, the item I2) that has been purchased by the visiting user from items co-occurring with the item I1 from the item set.

  When the item set is derived, the control unit 11 determines whether or not the item set element is empty (step S106). If it is determined that the item set is empty (Y in step S106), the control unit 11 proceeds to the process in step S102.

  On the other hand, when determining that the item set is not empty (for example, determining that the element of the item set R (I1) is [I2, I3, I4, I5]) (N in step S106), the control unit 11 One of the sets is identified as a co-occurrence item “r” for the user purchase item “t” (ie, one of the elements of the item set R (I1) is substituted for “r”) (step S107). ). The element selection method here is not particularly limited. For example, when the item ID is indicated by a numerical value, the element having the smallest numerical value may be selected.

  For example, when the element “I2” is substituted for “r”, the control unit 11 deletes the substituted element “I2” from the item set R (I1) (step S108). That is, hereinafter, the elements of the item set R (I1) are [I3, I4, I5].

  When the substituted element is deleted from the item set, the control unit 11 derives the correlation degree between the items assigned to the user purchase item “t” and the co-occurrence item “r” by the correlation degree deriving unit 14 (Step S11). S109).

  Here, the correlation degree deriving unit 14 calculates the correlation degree between the user purchase item “I1” and the co-occurrence item “I2” based on Equation 1 and purchase history information (see FIG. 3). Specifically, the correlation degree deriving unit 14 refers to the purchase history information DB 10A, the number of users who purchased each of the item I1 and the item I2 (including the number of users who have purchased both), the item I1 and the item The number of users who purchased both of I2 is specified, and the degree of correlation between item I1 and item I2 is calculated using Equation 1.

  If the correlation degree between items is derived | led-out, the control part 11 will transfer to the process of step S106.

  When the process from step S102 to S109 is repeated until the element of the input matrix T (U2) becomes empty, and it is determined that the element of the input matrix T (U2) is empty (Y in step S102), the control unit 11 The recommended item specifying unit 15 sorts the items in descending order of the derived correlation degree, and specifies a predetermined number of items (for example, three items having a high correlation degree) excluding the items already purchased by the visiting user U2 as recommended items. (Step S110).

  FIG. 8 is an explanatory diagram illustrating an example of the degree of correlation derived by the recommended item specifying process in this example. FIG. 8A is an explanatory diagram showing the degree of correlation for each element of the input matrix T (U2) derived by the control unit 11 in the processing of steps S101 to S109. As shown in FIG. 8A, when the correlation degree of each co-occurrence item with respect to the visiting user relation item [I1, I3] is derived, the control unit 11 calculates the sum of the derived correlation degrees (hereinafter referred to as “correlation degree”). (Recommendation level) is calculated (if the recommendation level is the same, for example, the ranking may be the same, or the ranking may be determined with reference to other elements such as the item ID) (see FIG. 8B). ). Next, the control unit 11 sorts the items in descending order of the recommendation level, and assigns a recommendation rank to the items except the items (items I1 and I3) that have been purchased by the visiting user U2 in descending order of recommendation level. (See FIG. 8C). When the recommendation order is given, the control unit 11 selects items in descending order of recommendation order according to the number of recommended items stored in the storage area included in the recommended item search server 10 in advance, and uses the selected items as recommended items. Identify. Hereinafter, the case where the control unit 11 specifies the items I2, I4, and I5 as recommended items by the recommended item specifying unit 15 will be described as an example.

  When the recommended item is specified, the control unit 11 creates a recommended item display screen including the recommended item by the item recommendation unit 16 (that is, creates data for displaying the recommended item display screen) (step S111).

  When the recommended item display screen is created, the control unit 11 transmits the created recommended item display screen to the user terminal 22 (step S112), and ends the processing here.

  FIG. 9 is an explanatory diagram illustrating an example of a recommended item display screen displayed on the display screen of the display device of the user terminal 22. As shown in FIG. 9, the recommended item display screen includes a recommended item display area 201 that displays selectable items 201 </ b> A to 201 </ b> C in a selectable manner, and a return button 202 that receives a display request for an item search screen. For example, when the recommended item search server 10 receives a selection notification of the recommended item 201A from the user terminal 22, the control unit 11 displays a detailed information display screen for displaying detailed information of the item I2 displayed as the item 201A. Is transmitted to the user terminal 22.

  In the above-described embodiment, the case has been described where the input matrix is specified based on the visitor's purchase history and the item set corresponding to the specified input matrix is specified in the recommended item specifying process. However, the function of the recommended item search server 10 is not limited to this. In other words, for example, an item that a visiting user plans to purchase may be specified as a visiting user related item. With this configuration, it is possible to provide recommended items even for users who do not have a purchase history. Hereinafter, the operation of the recommended item search server 10 in this case will be described with reference to the drawings. Note that the contents of operations and processes not particularly related to the present invention may be omitted.

  FIG. 10 is a flowchart illustrating an example of recommended item specifying processing executed by the recommended item search server 10 in this case. Here, the description will focus on the difference from the above-described recommended item specifying process, and the same step number is assigned to the same process as the above-described recommended item specifying process, and the description thereof is omitted.

  When the recommended item is identified from the items that the visiting user plans to purchase, the control unit 11 first selects an item (scheduled purchase item) that the visiting user plans to purchase in the recommended item identifying process. It is specified as “user relation item” (step S101A). It should be noted that the “item to be purchased” means that the recommended item search server 10 determines whether or not the “visiting user” is in accordance with a predetermined determination condition regardless of whether or not the visiting user is actually planning to purchase. Means an item that has been determined to be purchased.

  In this example, the control part 11 specifies the item currently displayed on the display screen of the display apparatus with which the user terminal 22 which a visiting user operates is a purchase plan item. Note that, for example, the determination condition as to whether or not the purchase is planned may be configured to include, for example, the cumulative display count and display time of the item.

  When the visiting user-related item is specified, the control unit 11 specifies the specified user-related item as “t” (user purchase item) in Formula 1 (that is, the visiting user-related item specified in “t” in Formula 1). (Step S103A).

  Next, the control unit 11 uses the item set deriving unit 13 to identify the item set R (t) co-occurring with “t” (step S105). In addition, about the specific method, since the method mentioned above can be used (refer FIG. 7 (B), (C)), description here is abbreviate | omitted.

  When the item set is specified, the control unit 11 determines whether or not the element of the specified item set is empty (step S106A). If it is determined that the element of the identified item set is not empty (N in step S106A), the control unit 11 performs the same processing as in steps S107 to S109 described above, and proceeds to the processing in step S106A.

  On the other hand, if it is determined that the element of the identified item set is empty (Y in step S106A), the control unit 11 performs the same processing as in steps S110 to S112 described above.

  Note that FIG. 11 is displayed on the display screen of the display device of the user terminal 22 that has received the recommended item display screen created by the recommended item search server 10 when the recommended item is specified from the planned purchase items of the visiting user. It is explanatory drawing which shows the example of an item browsing screen. As shown in FIG. 11, the item browsing screen in this example is recommended based on an item information display area 301 for displaying various information (for example, item name and price) related to the planned purchase item and the received recommended item display screen. A recommended item display area 302 for selectively displaying the item 302A and a close button 303 for accepting a non-display request for the recommended item display area are included. For example, when the recommended item search server 10 receives a selection notification of the recommended item 302A from the user terminal 22, the control unit 11 displays the detailed information of the item I2 displayed as the item 302A. Is transmitted to the user terminal 22.

  In addition, when it is set as the structure which displays a some item on the display screen of the display apparatus with which the user terminal 22 which a visiting user operates, or the item displayed on the display screen within the predetermined period is preserve | saved temporarily, and a purchase plan In the case of a configuration to handle as an item, the control unit 11 uses a display screen of the user terminal 22 in place of the purchase history of the visiting user in the recommended item specifying process when specifying the input matrix from the purchase history of the visiting user. What is necessary is just to set it as the structure which handles the some item currently displayed or the purchase plan item group temporarily preserve | saved as the input matrix T (refer FIG. 5).

  As described above, in the embodiment described above, based on the purchase history information indicating the item purchase history of each user, the recommended item that identifies the recommended item recommended for purchase to the visiting user who visited the purchase site. The search server 10 includes a purchase history information DB 10A for storing purchase history information, refers to the purchase history information, and is purchased by an item (for example, items I1, I3) purchased by a visiting user (for example, user U2) or by a visiting user. The related user (for example, the related user RU (t)) that is another user who has a purchase history of the visiting user related item indicating the item scheduled for the purchase is identified, the purchase history information is referenced, and the identified related user is Deriving an item set (for example, item set R (t)) that is a set of purchased items, Since the degree of correlation between each item belonging to the derived item set is derived, and according to the derived degree of correlation, a predetermined number of items having a high degree of correlation with the visiting user relationship item are specified as recommended items. The recommended items can be specified in real time without requiring a large-capacity memory.

  That is, for each recommendation request from the purchase history information, the recommended item is specified by directly calculating from the input data (for example, the visitor related item) (for example, the purchase history information includes the correlation degree of other items with respect to the visitor related item). Therefore, only the original matrix (for example, the purchase history matrix table, see FIG. 3) needs to be stored in the DB provided in the recommended item search server 10 to save space (storage area). Will be able to.

  Also, the time for preparing a so-called neighborhood matrix (neighbor information, see FIG. 4) in advance becomes “0”, and it becomes possible to deal with changes in the input matrix (for example, the input matrix T) in real time.

  In addition, regarding the derivation of the degree of correlation, only the degree of correlation between the item purchased by the visiting user and its co-occurrence item (for example, the degree of correlation based on the cosine distance, etc.) is obtained, so the recommended item can be identified with the minimum calculation cost Will be able to.

  Moreover, it can be said that the recommended item search process (refer FIG. 5) which the recommended item search server 10 in embodiment mentioned above performs only does not depend on a matrix size but only a matrix density, if compared with a naive method, for example. . Therefore, it can be said that the recommended item search system 100 including the recommended item search server 10 is scalable.

  Here, “matrix size” means a matrix area, that is, a product of the number of rows and the number of columns in one matrix. In addition, “matrix density” means the proportion of elements in the matrix where “0” is not set (that is, “1” is set for the number of elements of the matrix in this example). Means the ratio of the number of elements.) In general, when the purchase history is represented by a matrix, the matrix density of the matrix (purchase matrix) is only about 0.01% to 0.1%. For example, a matrix of 100 × 100 (a matrix of area 10,000) If the density is 0.1%, the number of elements for which “1” is set (in the sense of “flag”, the number of elements for which “1” stands) is “10 (pieces)”. Therefore, the number of elements that must be referred to when calculating the degree of correlation can be greatly reduced compared to the matrix area. In the embodiment described above, the recommended item search server 10 stores the information indicating the position where “1” stands in the purchase matrix (see FIG. 3) in the storage area. Since the calculation is executed only for the position where "" stands, the processing load required for calculating the degree of correlation increases in proportion to the matrix area expansion rate (that is, the rate of increase in the number of items and the number of users). There is no extensibility. Therefore, it can be said that the recommended item search system in the above-described embodiment is “scalable without depending on the matrix area”. Note that the processing load required to calculate the degree of correlation is proportional to the matrix density.

  In the embodiment described above, the purchase history information is stored in a purchase history matrix table having an item axis in which each item is arranged and a user axis in which each user is arranged. Since the purchase record information indicating the presence / absence of purchase record is stored in each cell (see FIG. 3), the purchase history can be efficiently held.

  Moreover, in embodiment mentioned above, the recommendation item search server 10 confirms each square corresponding to the visit user related item (for example, item I1) in the item axis | shaft of a purchase history matrix table, and the purchase track record of a visit user related item is. Since the user (for example, the users U1, U3, U4, U5) on the user axis corresponding to each square in which the purchase record information indicating the presence is stored is specified as a related user (see FIG. 7), it is simple. With this process, the related user corresponding to the visiting user related item can be specified.

  In the embodiment described above, the recommended item search server 10 confirms each cell corresponding to the related user in the user axis of the purchase history matrix table, and an item set (for example, item set) that is a set of items purchased by the related user Since the set R (t)) is derived (see FIG. 7), the set of items purchased by the related user can be derived by a simple process.

  In the above-described embodiment, the recommended item search server 10 recommends the specified recommended item to the visiting user (for example, a recommendation for displaying the name of the recommended item on the display screen of the display device included in the user terminal). Since the data for displaying the display screen including the item display area is created and the created data is transmitted to the user terminal), the user can easily recognize the recommended item.

  Although not specifically mentioned in the above-described embodiment, the recommended item search server 10 can identify the recommended item and the visiting user in response to the recommendation of the item to the visiting user. (For example, item ID and user ID) and information (recommendation information) in which the date and time of recommending the item are associated with each other are temporarily stored in the storage area of the storage unit 20, and an item recommendation request from the user terminal is made. When received, the recommendation information is referred to, and when it is determined that the date and time when the recommendation request is received has not passed a predetermined time (for example, 24 hours) from the date and time when the item indicated by the recommendation information is recommended It is good also as a structure which specifies the item which item ID contained in is recommended item.

  With such a configuration, it is possible to realize a function of avoiding a recommendation calculation for a visited user for a certain period of time by caching the result for a visiting user who has made a recommendation once, and recommending items The processing burden on the search server can be reduced.

  The present invention is useful for constructing a system capable of specifying recommended items in real time without requiring a large-capacity memory.

10 Recommended Item Search Server 10A Purchase History Information DB
DESCRIPTION OF SYMBOLS 11 Control part 12 Relation user specific part 13 Item set derivation part 14 Correlation degree derivation part 15 Recommended item specific part 16 Item recommendation part 21-2N User terminal 30 Base station 40 Web front end server (WFE server)
50 Communication Network 100 Recommended Item Search System

Claims (6)

A recommended item search server for identifying a recommended item recommended for purchase to a visiting user who visited a purchase site based on purchase history information indicating an item purchase history of each user,
Purchase history information storage means for storing the purchase history information;
A related user who refers to the purchase history information and identifies a related user who is another user who has a purchase history of a visiting user related item indicating an item purchased by the visiting user or an item planned to be purchased by the visiting user. Specific means,
Item set derivation means for referring to the purchase history information and deriving an item set that is a set of items purchased by the related user specified by the related user specifying means;
Correlation degree deriving means for deriving a degree of correlation between the visiting user-related item and each item belonging to the item set derived by the item set deriving means;
A recommendation item specifying means for specifying, as the recommended item, a predetermined number of items having a high degree of correlation with the visiting user-related item according to the correlation degree derived by the correlation degree deriving means. Item search server. The purchase history information is stored in a purchase history matrix table having an item axis in which each item is arranged and a user axis in which each user is arranged,
The recommended item search server according to claim 1, wherein purchase record information indicating the presence / absence of purchase record is stored in each square in the purchase history matrix table. The related user specifying means confirms each square corresponding to the visiting user related item in the item axis of the purchase history matrix table, and stores purchase result information indicating that there is a purchase record of the visited user related item. The recommended item search server according to claim 2, wherein a user on a user axis corresponding to each square is specified as a related user. The item set deriving unit confirms each cell corresponding to the related user on the user axis of the purchase history matrix table, and derives an item set that is a set of items purchased by the related user. 3. The recommended item search server according to 3. The recommended item search server according to any one of claims 1 to 4, further comprising item recommendation means for recommending the recommended item specified by the recommended item specifying means to the visiting user. A recommended item search program for identifying recommended items recommended for purchase by visiting users who have visited a purchase site based on purchase history information indicating an item purchase history of each user,
On the computer,
With reference to the purchase history information stored in the purchase history information storage means for storing the purchase history information, a visitor-related item indicating an item purchased by the visiting user or an item that the visiting user plans to purchase A related user identification step for identifying a related user who is another user having a purchase history;
An item set deriving step for deriving an item set that is a set of items purchased by the related user identified in the related user specifying step with reference to the purchase history information;
A degree of correlation deriving step for deriving a degree of correlation between the visiting user-related item and each item belonging to the item set derived in the item set deriving step;
A recommended item search for executing a recommended item specifying step of specifying, as the recommended item, a predetermined number of items having a high degree of correlation with the visiting user-related item according to the correlation level derived in the correlation level deriving step program.

Images