JP2006190164A - Information processing apparatus and method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To properly decide a recommendation item by considering the property of an item. <P>SOLUTION: An operation history management part 11 generates a user liking vector individually by utilizing the past operation history of a user relative to each of one or more items having corresponding character by each of a plurality of characters which the item can take, and stores the user liking vector in a UPV/UPV management table storage part 12. A recommendation list generation part 14 decides the character of a recommendation object from among the plurality of the characters, obtains the user liking vector about the character of the recommendation object from the UPV/UPV management table storage part 12, obtains the feature vector of each of the one or more items having the character of the recommendation object from an item feature vector storage part 13, calculates similarity of them, and decides the item corresponding to the feature vector where the calculated similarity satisfies a prescribed condition as the recommendation item. This invention can be applied to a recommendation item for recommending an item such as wine. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus and method, and a program, and in particular, an information processing apparatus that can appropriately determine a recommended item to be recommended to a user by considering each property of items with different user preference trends. And a method and a program.

  In recent years, a system for recommending items to users (hereinafter simply referred to as a recommendation system) has become widespread as one of information processing systems (see, for example, Patent Document 1). An item refers to, for example, a product that can be purchased by a user, and is described as content in Patent Document 1. Details of the item will be described later.

In this conventional recommendation system, one vector expressing the user's preference (hereinafter referred to as a user preference vector) and one vector expressing an item as a candidate for recommendation (hereinafter referred to as an item feature vector) And the comparison result is expressed with a certain scale (distance, angle, etc.), and the scale is a similarity, and when the similarity satisfies a predetermined condition, the item corresponding to the item feature vector is It was determined as a recommended item to be recommended to the user.
JP 2004-194108 A

  However, the conventional recommendation system including Patent Literature 1 has a problem that even if items of different properties such as types and fields are recommended, the difference in user preference for each property cannot be sufficiently considered. there were.

  In other words, an item that does not have the property that the user likes or an item that has the property but is not appropriate for the user (an item with a different preference tendency) is presented to the user as a recommended item. There was a problem that sometimes. In addition, when multiple recommended items are presented to the user across multiple types of properties, it is very difficult to optimize or personalize the distribution ratio of the number of recommended items presented for each type of property. There was a problem that there was.

  In the recommendation system described in Patent Document 1, positive history vectors and negative history vectors are used as user preference vectors. However, the positive history vector and the negative history vector are classified into two types of positive and negative, ie, the history of the item that the user has operated in the past, the history of the operation itself, etc. A vector generated for each type. Therefore, for example, when positive / negative is defined as likes / dislikes, it is determined whether a predetermined item is similar to an item that the user has determined to like in the past or similar to an item determined to be disliked. For example, a positive history vector and a negative history vector are used. As described above, the positive history vector and the negative history vector cannot be said to be a vector in consideration of the property of the item.

  Patent Document 1 suggests that a recommended item is determined using a positive history vector or a negative history vector for each genre of content (television program) or for each performer.

  However, the positive history vector or negative history vector for each genre or performer alone does not necessarily reflect the user's preference tendency for each genre or each performer. When a recommended item is determined using each positive history vector or negative history vector, the recommended item is necessarily an appropriate item for the user, even if the recommended item has the property that the user likes itself. However, the above-described problem that it is not limited (there may be a recommended item different from the preference tendency) occurs as it is.

  Further, when a plurality of recommended contents across a plurality of genres and a plurality of performers are presented to the user, the patent document 1 naturally suggests how to determine the distribution ratio of the number of presentations and the like. It has not been done. Therefore, in the recommendation system described in Patent Literature 1, it is very difficult to optimize or personalize the distribution ratio of the number of presentations of recommended content for each genre or for each performer. It just happens.

  As described above, the recommendation system described in Patent Document 1 cannot sufficiently consider the difference in the user's preference tendency for each property.

  The present invention has been made in view of such a situation, and it is possible to appropriately determine a recommended item to be recommended to a user by considering each property of an item having a different preference tendency by the user. To do. Note that “determined appropriately” here refers to recommending a plurality of types of properties in addition to appropriately determining recommended items having that type of property as a recommendation target. In addition to appropriately determining the recommended item for each type of property as a target, it also includes appropriately determining the ratio of the number of presentations for each type of property.

  The information processing apparatus according to the present invention uses, for each of a plurality of properties that an item can take, a user preference vector that represents a user's preference using a user's past operation history for each of one or more items having the corresponding property. About the preference vector generation means to be generated individually, the property of the recommendation target from among the plurality of properties, and the property of the recommendation target in the user preference vector for each of the plurality of properties generated by the preference vector generation means The similarity between the user preference vector and the feature vector indicating the feature of each of the one or more items having the property to be recommended is calculated, and the item corresponding to the feature vector in which the calculated similarity satisfies the predetermined condition And recommending means for determining as a recommended item to be recommended to the user.

  Storage means for storing the user operation history is further provided, and the preference vector generation means further performs control for storing the user operation history in the storage means in a form that can be managed individually for each of a plurality of properties. it can.

  The recommendation unit may determine two or more of the plurality of properties as the properties of the recommendation target, and individually determine one or more recommended items for each of the two or more properties of the recommendation target.

  When at least a part of the recommended items determined for each property of the recommendation target by the recommendation means is presented to the user, the operation history stored in the storage means is used for each of the properties of the recommendation target It is possible to further provide a distribution determining means for determining the distribution ratio of the presentation of the.

  The information processing method of the present invention uses, for each of a plurality of properties that an item can take, a user preference vector that represents a user's preference using a user's past operation history for each of one or more items having a corresponding property. A preference vector generation step that is generated individually, a property of a recommendation target is determined from a plurality of properties, and a recommendation target of user preference vectors for each of the plurality of properties generated by the processing of the preference vector generation step The similarity between the user preference vector for the property and the feature vector indicating each feature of the one or more items having the property to be recommended is calculated, and the calculated similarity corresponds to the feature vector satisfying a predetermined condition. A recommendation step of determining an item as a recommended item to be recommended to the user.

  The program of the present invention is a program that is executed by a computer that controls a recommendation device that recommends an item to a user, and relates to each of one or more items having a corresponding property for each of a plurality of properties that the item can take. A preference vector generation step for individually generating a user preference vector representing the user's preference using a user's past operation history, and a property of a recommendation target is determined from a plurality of properties, and a preference vector generation step is performed The similarity between the user preference vector for the property of the recommendation target among the user preference vectors for each of the plurality of properties generated by the feature vector and the feature vector indicating the feature of each of the one or more items having the property of the recommendation target The item corresponding to the feature vector whose calculated similarity satisfies the predetermined condition , Characterized in that it comprises a recommendation determining as recommendation items to be recommended to the user.

  In the information processing apparatus, method, and program of the present invention, for each of a plurality of properties that an item can take, the user's preference is obtained by using the user's past operation history for each of the one or more items having the corresponding property. A user preference vector representing is generated individually. Then, the property of the recommendation target is determined from the plurality of properties, and the user preference vector for the property of the recommendation target among the user preference vectors for each of the plurality of properties generated by the processing of the preference vector generation step; The similarity with the feature vector indicating each feature of one or more items having the property of the recommendation target is calculated, and the item corresponding to the feature vector that satisfies the predetermined condition should be recommended to the user It is determined as a recommended item.

  As described above, according to the present invention, an item can be recommended to a user. In particular, it is possible to appropriately determine a recommended item to be recommended to the user in consideration of the type and field of the item.

  Embodiments of the present invention will be described below. Correspondences between constituent elements described in the claims and specific examples in the embodiments of the present invention are exemplified as follows. This description is to confirm that specific examples supporting the invention described in the claims are described in the embodiments of the invention. Therefore, even though there are specific examples that are described in the embodiment of the invention but are not described here as corresponding to the configuration requirements, the specific examples are not included in the configuration. It does not mean that it does not correspond to a requirement. On the contrary, even if a specific example is described here as corresponding to a configuration requirement, this means that the specific example does not correspond to a configuration requirement other than the configuration requirement. not.

  Further, this description does not mean that all the inventions corresponding to the specific examples described in the embodiments of the invention are described in the claims. In other words, this description is an invention corresponding to the specific example described in the embodiment of the invention, and the existence of an invention not described in the claims of this application, that is, in the future, a divisional application will be made. Nor does it deny the existence of an invention added by amendment.

  According to the present invention, an information processing apparatus is provided. This information processing apparatus (for example, the recommendation server 1 in FIG. 1) is provided for each of a plurality of properties that an item can take (for example, when the item is wine, red wine, white wine, and sparkling wine, as will be described later). Preference vector generating means for individually generating a user preference vector representing the user's preference using the user's past operation history for each of one or more items having corresponding properties (for example, operation history management in FIG. 1) Part 11) and a property of a recommendation target among the plurality of properties, and the property of the recommendation target among the user preference vectors for each of the properties generated by the preference vector generation means A user preference vector, and a feature vector indicating each feature of one or more items having the property to be recommended Recommendation means for calculating similarity and determining an item corresponding to the feature vector satisfying a predetermined condition for the calculated similarity as a recommended item to be recommended to the user (for example, the recommendation list generator 14 in FIG. 1). A UPV selection unit 41 and a recommendation selection unit 42).

  The information processing apparatus further includes a storage unit (for example, the UPV / UPV management table storage unit 12 in FIG. 1) that stores the operation history of the user, and the preference vector generation unit stores the operation history of the user. Further, it is possible to further perform control for storing in the storage means in a form that can be managed individually for each of the plurality of properties (for example, a form like the UPV management table in FIGS. 2 to 4).

  The recommendation unit of the information processing apparatus may include two or more of the plurality of properties (for example, when the item is wine, red wine and white wine among red wine, white wine, and sparkling wine as will be described later). ) As the property of the recommendation object, and one or more recommended items are individually determined for each of the properties of the two or more recommendation objects.

  When at least a part of the recommended items determined for each of the properties to be recommended by the recommendation unit is presented to the user, the information processing apparatus displays the operation history stored in the storage unit. Utilizing distribution determining means (for example, the display distribution determining unit 43 of the recommendation list generating unit 14 in FIG. 1) for determining the distribution ratio (presentation ratio to be described later) of each of the properties to be recommended. It can be provided.

  According to the present invention, an information processing method is provided. In this information processing method, for each of a plurality of properties that an item can take, a user preference vector that represents the user's preference is individually obtained using a user's past operation history for each of one or more items having the corresponding property. A preference vector generation step (for example, the process of step S2 of FIG. 5) to be generated, a property to be recommended is determined from the plurality of the properties, and the plurality of properties generated by the processing of the preference vector generation step Calculating a similarity between a user preference vector for the property of the recommendation target among the user preference vectors for each of the feature vectors and a feature vector indicating each feature of the one or more items having the property of the recommendation target, An item corresponding to the feature vector whose calculated similarity satisfies a predetermined condition should be recommended to the user. Recommendation determining as an item (e.g., "recommendation list generation process" in step S4 of FIG. 5 and FIG. 6); and a.

  Furthermore, according to the present invention, there are also provided a program corresponding to the information processing method of the present invention described above and a recording medium on which the program is recorded. As will be described later, this program is executed by, for example, the computer shown in FIG.

  Next, the principle of the present invention will be described before describing the embodiment of the present invention.

  In the present specification, an item refers to software or hardware (article) that can be provided to a user. For example, software such as so-called content, that is, a television broadcast program, a movie, a photograph, a song, and the like (moving image, still image, sound, or a combination thereof) is also an item in this specification. Further, for example, an article (hardware) such as wine is also an item referred to in this specification. In addition, for example, sentences and conversations are also items in this specification.

  However, in order to facilitate understanding of the present invention, in the following, when it is necessary to give specific examples of items, wine (sake) packed in a predetermined container (for example, a 750 ml bottle) is adopted. Suppose that

  First, an outline of an example of a series of processes (hereinafter simply referred to as a recommendation process) until an item is generally recommended that has been conventionally performed will be described.

  However, for simplicity of explanation, it is assumed that one information processing apparatus executes all of the recommendation processing.

  First, a conventional information processing apparatus vectorizes an item by using each of N pieces (N is an integer value of 1 or more) of information indicating the characteristics of an item as a base vector. At that time, the conventional information processing apparatus weights each component of the vector using a predetermined weighting method (gives a weight value) as necessary.

  A vector obtained as a result of vectorizing items in this manner is the item feature vector described above. Hereinafter, information indicating the feature of an item is referred to as attribute information, and the type of attribute information is referred to as an attribute. That is, when attribute information of N attributes is associated with a predetermined item, N attribute information is digitized (weighted), and N numerical values obtained as a result are obtained. Is the item feature vector.

  When attribute information is not associated with an item for a predetermined attribute of N attributes, 0 is assigned to the component corresponding to the predetermined attribute of the item feature vector of the item. Let's say.

  A conventional information processing apparatus generates and accumulates such item feature vectors for each item. That is, an item corresponding to each item feature vector accumulated in this manner is a candidate for recommendation.

  On the other hand, a conventional information processing apparatus generates a vector indicating a user's preference using a user's past operation history and various information input by the user himself / herself. This vector is the above-described user preference vector. Hereinafter, the user preference vector is appropriately referred to as UPV (User Preference Vector).

  Then, the conventional information processing apparatus obtains a similarity such as cosine correlation between the UPV and each of the accumulated item feature vectors (matching process is performed), for example, the similarity is equal to or greater than a threshold value. An item corresponding to the item feature vector is presented to the user as a recommended item to be recommended to the user.

  The outline of the recommendation process in the conventional recommendation system has been described above.

  By the way, when multiple items that are candidates for recommendation span multiple types and fields, even if they are the same user, the user's preference tendency will naturally change according to the type and field of items. There are many.

  For example, when the item is wine, wine can be classified into three types: red wine, white wine, and sparkling wine. More precisely, since there are other types such as rose wine (which may be included in red wine) and refined wine, wine is actually classified into more than three types, but here, For simplicity of explanation, only three types are mentioned. In this case, generally, the number of users showing the same preference tendency is small for all three types of red wine, white wine, and sparkling wine, and the number of users showing different preference trends for each of the three types is large.

  Specifically, for example, even for the same user, the preference tendency is “if red wine, full-bodied astringent burgundy”, “white wine, sweet German wine”, and “sparkling wine” In many cases, they tend to be completely different, such as “dry champagne”. That is, focusing only on the attribute of “taste”, the user's preference tendency is “full-bodied astringency” for red wine, “sweet” for white wine, and “dry” for sparkling wine. "There are three types of red wine, white wine, and sparkling wine. Similarly, focusing only on the attribute of “origin”, the red wine is in the “Burgundy” region of France, the white wine is in “Germany”, and the sparkling wine is in the “Champagne” region of France. There are three types of wines: red wine, white wine, and sparkling wine.

  However, in the case of recommending wine in the conventional recommendation system, each of the three types of red wine, white wine, and sparkling wine, although the same user shows completely different preference trends, these three types are recommended. The conventional problem mentioned above that the difference of the user's preference about the kind was not considered arises. That is, in the conventional recommendation system, a user preference vector for mere “wine”, which is a combination of these three types, is generated, and the recommended wine is simply determined using the user preference vector. The conventional problem of arises. Therefore, it is obvious that there is a problem that the recommended wine determined in this way is not necessarily appropriate wine for the user.

  Therefore, the present inventors have invented the following method.

  For example, when recommending wine, a user preference vector for red wine, a user preference vector for white wine, and a user preference vector for sparkling wine are generated and stored and managed. Then, if necessary, the similarity between the user preference vector for red wine and the item feature vector of red wine is calculated, and the red wine corresponding to the item feature vector that satisfies the predetermined condition is determined as a recommended item. To do. Similarly, if necessary, the similarity between the user preference vector for white wine and the item feature vector of white wine is calculated, and the white wine corresponding to the item feature vector whose similarity satisfies a predetermined condition, Determine as a recommended item. If necessary, the similarity between the user preference vector for sparkling wine and the item feature vector of the sparkling wine is calculated, and the sparkling wine corresponding to the feature vector that satisfies the predetermined condition is determined as the recommended item. To do. A technique corresponding to the series of processes described above is a technique newly invented by the present inventors.

  In other words, the technique of the present invention is a technique capable of realizing the following first process and second process after all.

  That is, the first process generates each of the user preference vectors by using the past operation history of the user regarding each of one or more items having the corresponding property for each of a plurality of properties that the item can take. , And so on.

  Note that the past operation history of the user regarding the item having the corresponding property is, for example, an operation performed by the user using the information processing apparatus when the item is wine and the corresponding property is red wine. Information on the operation of the specified red wine (for example, red wine named Chateau XX), the input operation of the tasting result (evaluation) of the chat XX, and the information on the chat XX on the Internet etc. This refers to operations until the acquisition, or various operations described on the left of the whole red wine, not a specific one red wine such as Chateau XX. Furthermore, when the item is not an article such as wine but software that can be used using an information processing device such as content, the operation of the software itself also relates to an item having a corresponding property. It is an example of a user's past operation history.

  Further, as described above, Patent Document 1 describes generating a positive history vector and a negative history vector for each genre and performer, but a positive history vector and a negative history vector for each genre and performer. Does not necessarily take into account (does not reflect) the difference in the preference tendency of each property by the user. That is, Patent Document 1 proposes to generate a positive history vector and a negative history vector for each genre or performer simply because there is a like or dislike of the genre itself or the performer itself. Therefore, for example, in the drama positive history vector generated according to the method of Patent Document 1, as described in Patent Document 1, even if it is not a particularly favorite actor, it simply appears frequently in the drama. The point (component value) of performer B becomes a high point. However, even if you say that you like dramas, the reason (favorite tendency) is because it is not because performer B frequently appears in dramas but for other reasons. Is not necessarily reflected in the drama positive history vector. Therefore, simply referring to Patent Document 1, individual characteristics are considered in consideration of differences in the user's preference tendency, such as the three types of characteristics of red wine, white wine, and sparkling wine. It is not easy to think about generating user preference vectors. In other words, the item property referred to here is not just a generally classified property, but a property that has been reclassified or selected in consideration of a difference in user preference. That is, as described later, there are various other properties such as the production area and price, as described later, but it is not necessary to generate user preference vectors individually for all of these properties. The user preference vector may be generated for each property in which the difference in the user's preference tendency is considered (the difference becomes clear).

  The second process is to determine the property of the recommendation target among the properties that the item can take, and the user preference vector for the property of the recommendation target and the item feature vector of one or more items having the property of the recommendation target This is a process of calculating the degree of similarity with each item, and determining an item corresponding to the item feature vector that satisfies the predetermined condition for the calculated degree of similarity as a recommended item having a property to be recommended.

  When a plurality of properties are determined as the recommendation targets, one or more recommended items are determined for each of the plurality of properties that are recommended targets as a result of the second process. For example, when the item is wine, and all of red wine, white wine, and sparkling wine are recommended, the recommended item (recommended wine) is one or more red wines, one or more white wines, and One or more sparkling wines will be determined respectively. In this case, it is naturally possible to present all the recommended items to the user, but when the total number is too large, or when the display screen for presenting the recommended items as images to the user is small (for example, Etc.) when the user is using a portable terminal.

  Therefore, in such a case, it is necessary to select some recommended items from all recommended items and present them to the user. The selection method at this time is not particularly limited, and the recommended item having the highest recommendation level (m is an arbitrary number of 1 or more) is uniformly selected for all of the plurality of properties to be recommended. A technique, that is, a technique in which the ratio of the number of presentations is made uniform for all properties of the recommendation target may be adopted. However, for a user, the plurality of properties that are targeted for recommendation do not all have the same importance, and when this method is adopted, the ratio of the number of presentations is not necessarily appropriate for the user. Problems arise.

  Therefore, in order to solve this new problem, the present inventor determines the ratio of the number of presentations according to the first condition when presenting recommended items to the user in addition to the above-described method of the present invention. Also invented. When this method is applied, for each property to be recommended, a number determined by the ratio of the number of presentations for the corresponding property and the total number of presentations from among a plurality of recommended items having the corresponding property (this number Is an arbitrary integer value of 1 or more and is often a different value for each property), and each of the recommended items is selected according to the second condition. The recommended item can be presented to the user. In other words, it is possible to present recommended items to the user in various numbers for each property (since the distribution ratio is often different).

  The first condition is not particularly limited, and various conditions can be applied. Specifically, for example, when an item is wine and attention is paid to a purchase operation history as an operation history, for a user who purchases red wine, white wine, and sparkling wine at a ratio of 7: 2: 1. The condition that the ratio of the number of purchases (number of operation histories) is the ratio of the number of presentations can be applied as the first condition. In this case, the distribution ratio of the presentation is also 7: 2: 1. That is, the ratio of the number of presentations of red wine is 70%, the ratio of the number of presentations of white wine is 20%, and the ratio of the number of presentations of sparkling wine is 10%. Therefore, if ten recommended wines can be presented, each of seven red wines, two white wines, and one sparkling wine is presented to the user as a recommended wine. Other specific examples of the first condition will be described later with reference to FIGS.

  The second condition is also not particularly limited. For example, the number of presentations for a predetermined property is n (n is an arbitrary number greater than or equal to 0. However, n may be a different value for each property. If the number of recommended items has a predetermined property, the recommended item having the highest recommendation level up to the n-th recommended item may be selected.

  In addition, in the example mentioned above, since the determination timing of the ratio of the number of presentations was after the recommended item determination, for example, n recommended items were selected from a plurality of recommended items having a predetermined property. However, as a matter of course, the determination timing of the ratio of the number of presentations may be before the recommended item determination. In this case, if it is determined in advance that the number of presentations for a predetermined property is n, n recommended items are selected from a plurality of items (candidates for recommended items) having the predetermined property. Simply decide.

  The principle (method) of the present invention has been described above.

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

  FIG. 1 shows a functional configuration example of a recommendation system to which the above-described method of the present invention is applied, that is, a recommendation system to which the present invention is applied. That is, the recommendation system in FIG. 1 searches for recommended items to be recommended to the user from among a plurality of items, and presents the search results to the user.

  In the example of FIG. 1, the recommendation system includes a recommendation server 1, an application server 2, and user equipment 3.

  In addition, although each of the recommendation server 1, the application server 2, and the user equipment 3 is illustrated in the example of FIG. 1, the number of them is not limited to the example of FIG. . For example, when there are a plurality of users, there are user devices 3 used by each of the plurality of users, and naturally, the number of user devices 3 is not necessarily one.

  The recommendation server 1 communicates with the application server 2. Note that the form of communication is not particularly limited. For example, although not shown in FIG. 1, one or more networks may be interposed. The application server 2 communicates with the user device 3 in addition to the recommendation server 1. The form of communication is not particularly limited, and in the example of FIG. 1, the network 4 is interposed. However, the form of communication is not limited to the example of FIG. 1. For example, direct communication is performed without the network 4. The form to perform may be sufficient. Further, the form of the network 4 is not particularly limited. However, in the present embodiment, the network 4 is assumed to be the Internet, for example.

  The recommendation server 1 mainly executes processing corresponding to the above-described method of the present invention. For this reason, the recommendation server 1 is provided with an operation history management unit 11 to a recommendation list generation unit 14.

  When the user operates the operation unit 31 of the user device 3, the operation history management unit 11 acquires the operation content via the application server 2 and manages it as an operation history.

  What should be noted here is that the operation history management unit 11 individually manages the operation content for each property of the item. Specifically, for example, in the present embodiment, since the item is wine as described above, when attention is paid to the three types of red wine, white wine, and sparkling wine described above, the operation is performed. The history management unit 11 individually manages the operation content for red wine, the operation content for white wine, and the operation content for sparkling wine.

  The management method of the operation history management unit 11 is not particularly limited, but here, a method of generating and updating a user preference vector (UPV) for each property and generating and updating a UPV management table is adopted. And

  The operation history management unit 11 basically uses information (user operation history information) supplied from the operation history acquisition unit 21 of the application server 2 when generating or updating a UPV. Is stored in the item feature vector storage unit 13, for example, when the operation content such as the user viewing the detailed information of the specific wine or the operation content such as purchasing wine is included in this information. The item feature vector of the specific wine among the item feature vectors being used is also used.

  The UPV and UPV management table for each property are individually stored in the UPV / UPV management table storage unit 12 for each user.

  Specifically, for example, in the present embodiment, since the item is wine as described above, when attention is paid to the above-described three types of red wine, white wine, and sparkling wine, red wine The UPV for the white wine, the UPV for the white wine, and the UPV for the sparkling wine are stored and managed in the UPV / UPV management table storage unit 12.

  Further, the UPV management table stored and managed in the UPV / UPV management table storage unit 12 is a table used for determining the ratio of the number of recommended items presented above, and is individually used for each user. Refers to a table that is created and managed. In other words, as described above, the ratio of the number of presentations is determined according to the first condition, and a UPV management table appropriate for the first condition is stored and managed in the UPV / UPV management table storage unit 12. become.

  Specific examples of such UPV management tables are shown in FIGS. Therefore, specific examples of the UPV management table will be described individually in that order with reference to FIGS.

  FIG. 2 shows a specific example of the UPV management table used when the ratio of the number of presentations is determined according to the UPV update (access) frequency.

  That is, for example, the application server 2 now manages a home page (hereinafter referred to as a site) for providing (selling) services such as goods and contents (in this embodiment, wine) on the Internet 4. Suppose you are. Then, the user looks at this site (displays it on the display unit 32 of the user device 3), and performs an access operation (so-called click operation or the like) or purchase procedure operation for each item or a category to which the item belongs (operation unit 31). Perform the corresponding operation. In addition, each time this operation is performed, the UPV having the corresponding property is also updated. Note that this assumption also holds in the description of FIGS. 3 and 4 described later.

  In this case, it is possible to determine the ratio of the number of presentations according to the UPV update (access) frequency by accumulating the number of updates or the number of accesses for each UPV for each property. A specific example of a table in which the number of UPV updates or accesses is described is the UPV management table of FIG.

  In the UPV management table of FIG. 2, each row in the first column describes the name of the UPV in order from the top, and finally describes “total of each period”. What should be noted here is that each UPV described in each row of the first column is a UPV for a certain user. That is, a UPV for each property of an item for a single user is included in the UPV management table.

  In the following, for simplicity of explanation, it is assumed that “UPV1” is a UPV for red wine and “UPV2” is a UPV for white wine.

  Each column after the second column describes the number of updates or accesses of each UPV in each period, and the last column shows the number of updates or accesses of each UPV over the entire period. The number of times is described. Therefore, the column corresponding to the period including the current time and the column corresponding to the entire period are to be updated. Note that each period is not particularly limited, and may be any period, and does not need to have the same length as other periods. Furthermore, each period may span other periods. For example, “Period 1” may be one session (from the time the user opens the site until it is closed), and “Period 2” may be today (for example, from 6 am to 6 am the next day).

  For example, in “Period 1”, the number of times “UPV1” is updated or accessed, that is, the number of user operations for red wine is “3 times”, the number of times “UPV2” is updated or accessed, That is, it can be seen that the number of user operations for white wine was “one”. If “period 1” is a past period (eg, last week), the description content in the second column for “period 1” is not updated.

  Also, for example, in “Period 2”, the number of times “UPV1” has been updated or accessed, that is, the number of user operations for red wine is “10 times”, and the number of times “UPV2” has been updated or accessed That is, it can be seen that the number of user operations for white wine was “five times”. Note that when “period 2” is a period including the present time (for example, this week), the third column for “period 2” is an update target. That is, for example, when a user operation is subsequently performed on red wine, the description content in the second column of “UPV1” in the third column of “period 2” is changed from “10 times” to “11 times”. Will be updated. At that time, the description content of the last row of “total of each period” in the third column of “period 2” is updated from “18 times” to “19 times” and “all periods”. The content of the description in the second row of “UPV1” in the last column is updated from “58 times” to “59 times”.

  The UPV management table as in the example of FIG. 2 is stored in the UPV / UPV management table storage unit 12 and managed by the operation history management unit 11, so that a recommendation list generation unit 14 to be described later stores this UPV management table. It is possible to easily determine the ratio of the number of presentations according to the UPV update (access) frequency. A specific calculation method of the ratio of the number of presentations by the recommendation list generation unit 14 will be described later.

  FIG. 3 is used when the UPV update (access) frequency is weighted according to the access contents, and the ratio of the number of presentations is determined according to the weighted UPV update (access) frequency. A specific example of the UPV management table is shown.

  As described above, assuming that the same assumptions as in FIG. 2 are established in FIG. 3, the weight (load) corresponding to the access contents is integrated every time the UPV is updated or accessed, and the integration result , It is possible to determine the ratio of the number of presentations according to the weighted UPV update (access) frequency. A specific example of a table in which the integration result (the number of weighted UPV updates or the number of accesses) is described is the UPV management table in FIG.

  In the UPV management table of FIG. 3, each row in the first column describes the name of the UPV in order from the top, and finally describes “total”. In the following, for simplicity of explanation, “UPV1” is the same “UPV1” as in FIG. 2, that is, the UPV for the red wine of the same user, and “UPV2” is the same “UPV2” as in FIG. Suppose that it is UPV about white wine of the same user.

  In each of the second column to the fifth column, the number of times each UPV is updated or accessed in each of the “major category”, “small category”, “product details”, and “product purchase” is displayed. Described.

  Each of “Large Category”, “Small Category”, “Product Details”, and “Product Purchase” indicates access contents. That is, in general, the structure of a product sales site is hierarchized according to the size of a category, and a page for each level is often created. For this reason, when finding a purchase desired product, the user often specifies a purchase desired product by sequentially accessing a page of a small category from a page of a large category to narrow down the target. Therefore, access to the large category page is classified as “major category”, access to the small category page is classified as “small category”, and the detailed contents of each product including the purchase desired product are presented. Are classified as “product details”, and access associated with the purchase operation of the desired product is classified as “product purchase”.

  Specifically, for example, a wine sales site often has its production area as a category. For example, if a user wants to purchase a wine named Chateau XX in the Bordeaux region of France, for example, Many operations are performed. That is, for example, first, the user accesses a first page (for example, a top page) on which the name of each country is presented. For example, next, the user performs an operation of designating France from the first page, thereby accessing a second page on which names of French regions (Bordeaux, Burgundy, etc.) are presented. For example, next, the user performs an operation of designating Bordeaux from the second page to access a third page on which details (name, price, comment, etc.) of Bordeaux wine are presented. . For example, the user performs various operations necessary for purchasing the chatter XX in the third page. In this case, access to the first page is classified as “major category”, access to the second page is classified as “small category”, and access to the third page is classified as “product details”. In addition, access associated with various operations necessary for purchasing Chateau XX is classified as “product purchase”.

  In this case, the importance of access for the user generally increases in the order of “large category”, “small category”, “product details”, and “product purchase”. That is, when the user performs an operation up to “major category” and an operation up to “product details” for an item having a predetermined property, the latter case has a predetermined property. On the other hand, it can be considered that the user perceives it as more important (interested, etc.).

  Therefore, in the example of FIG. 3, for each of “large category”, “small category”, “product details”, and “product purchase”, the load (importance) is 1, 2, 3, 5 respectively. Is given. As a result, the UPV update (access) frequency can be weighted according to the access contents.

  In other words, each of the sixth columns, which are “accumulation points”, describes the number of updates or the number of accesses of each UPV after being weighted according to the access contents. Hereinafter, the number of times of updating or accessing each UPV after being weighted according to the contents of access is referred to as an “accumulation point” in accordance with the description of FIG.

  Specifically, for example, the “integrated point” of “UPV1” is “44” (= 3 (times) × 1 (load) +8 (times) × 2 (load) +5 (times) × 3 (load) +2 ( Times) × 5 (load)). Similarly, for example, the “integration point” of “UPV2” is “18” (= 1 (times) × 1 (load) +3 (times) × 2 (load) +2 (times) × 3 (load) +1 (times) × 5 (load)).

  The UPV management table as in the example of FIG. 3 is stored in the UPV / UPV management table storage unit 12 and managed by the operation history management unit 11, so that a recommendation list generation unit 14 to be described later stores this UPV management table. It is possible to easily determine the ratio of the number of presentations according to the “integrated points”. A specific calculation method of the ratio of the number of presentations by the recommendation list generation unit 14 will be described later.

  FIG. 4 shows a specific example of the UPV management table when the ratio of the number of presentations is determined according to the past operation tendency of the user.

  As described above, if the same assumptions as in FIG. 2 and FIG. 3 are established in FIG. 4, the UPV used for calculating the similarity is calculated by analyzing the tendency of the user's operation history. It is possible to calculate the transition probability of transition from the first UPV to the second UPV and the probability that two or more UPVs are simultaneously used for calculating the similarity, and determine the ratio of the number of presentations according to those probabilities It becomes possible. Thereby, for example, it becomes possible to follow the transition tendency (operation tendency) of a user's purchase appropriately and to recommend an item appropriately. A specific example of a table in which transition probabilities among such probabilities are described is the UPV management table in FIG.

  In the UPV management table of FIG. 4, each row in the first column describes the name of the UPV in order from the top, and finally describes “total”. In the following, for simplicity of explanation, “UPV1” is the same “UPV1” as in FIG. 2, that is, the UPV for the red wine of the same user, and “UPV2” is the same “UPV2” as in FIG. Suppose that it is UPV about white wine of the same user. Further, it is assumed that “UPV3” is a UPV for sparkling wine of the same user.

  In each of the second column and the third column, “integrated points” of each UPV are described for each of the short period and the long period.

  In each of the 4th to 6th columns, each UPV describes “transition probability to UPV1”, “transition probability to UPV2” from each UPV, and “transition probability to UPV3” from each UPV. Is done. Specifically, for example, focusing on the third row corresponding to “UPV2”, the “transition probability from“ UPV2 ”to“ UPV1 ”is“ 0.01 ”, and the“ transition probability from “UPV2” to “UPV2” is “0.24”, and the “transition probability from“ UPV2 ”to“ UPV3 ”is“ 0.45 ”. Therefore, since the transition probability from “UPV2” to “UPV3” is the highest, the user's operation tendency is to perform an operation on a sparkling wine (purchase operation) after an operation on white wine (purchase operation, etc.). I understand that there are many cases.

  The specific examples of the UPV management table have been described above with reference to FIGS. 2 to 4.

  Returning to FIG. 1, the item feature vector storage unit 13 stores an item feature vector for each of one or more items (wine in the present embodiment).

  The recommendation list generation unit 14 is configured to include a UPV selection unit 41, a recommendation selection unit 42, and a display distribution determination unit 43.

  Upon receiving a recommendation request from the recommendation acquisition unit 2 of the application server 2, the UPV selection unit 41 selects a UPV for a property to be recommended from each UPV (UPV for each property) of the user who is the presentation destination of the recommended wine. Are selected and acquired from the UPV / UPV management table storage unit 12. The method for determining the property of the recommendation target is not particularly limited, but here, for example, based on the content of the recommendation request and the current situation (user operation status, page of the site presented to the user, etc.) Assume that a method for determining the nature of the recommendation object is adopted.

  The recommendation selection unit 42 has a UPV regarding the property of the recommendation target selected by the UPV selection unit 41 and a property having a recommendation target among a plurality of item feature vectors stored in the item feature vector storage unit 13. Each similarity is obtained. Next, the recommendation selection part 42 calculates the recommendation degree of a corresponding item based on those similarities. The method of calculating the recommendation level is not particularly limited, but here, for simplicity of explanation, it is assumed that the similarity level is directly used as the recommendation level. Next, the recommendation selection unit 42 determines, for example, one or more items indicated by an item feature vector having a recommendation degree equal to or higher than a threshold value as a recommended item having a property to be recommended.

  In other words, the recommendation list generation unit 14 has a function of recommending an item appropriate for the user from items having the property of the recommendation target, using the UPV about the property of the recommendation target. In order to realize this function, the recommendation list generation unit 14 is provided with a UPV selection unit 41 and a recommendation selection unit 42.

  Further, the recommendation list generation unit 14 has a function of determining a presentation ratio for each property of the recommended item. In order to realize this function, the recommendation list generation unit 14 is further provided with a display distribution determination unit 43.

  That is, the display distribution determination unit 43 refers to the UPV management table stored in the UPV / UPV management table storage unit 12 and calculates (determines) the ratio of presentation for each of one or more properties to be recommended.

  Then, the recommendation selection unit 42 recommends the number based on the ratio of the presentation determined by the display distribution determination unit 43 for each of one or more properties to be recommended (this number is often different for each property). The items are selected, a list including each item for each selected property (hereinafter referred to as a recommendation list) is generated and supplied to the recommendation acquisition unit 23 of the application server 2.

  In practice, the recommendation list refers to a list in which information that can identify each of one or more recommended items is sorted in descending order of recommendation, for example. For example, in the case where the item is wine as in the present embodiment, the information that can identify the recommended item is an identifier (ID) unique to each wine, or the name, place of production, producer (domaine or Winery). It is also possible to include other information in the recommendation list, such as the price of wine and the reason for recommending that wine.

  By the way, although the calculation method of the ratio of presentation by the display distribution determination part 43 is not specifically limited, As mentioned above, it is necessary to be a calculation method which can utilize the UPV management table of a reference destination. Therefore, an example of a method for calculating the presentation ratio using the UPV management tables in the examples of FIGS. 2 to 4 will be described below.

  First, an example of a method for calculating the ratio of presentation that can use the UPV management table of the example of FIG. 2 will be described.

  That is, the target period is determined according to the recommended scene (page, etc.), in other words, the target column in the UPV management table of the example of FIG. 2 is determined, and the total number of periods in the period (the determined column) The display value for each UPV is divided by the number of updates (access) of each UPV (description value of the line indicating each UPV in the determined column). A calculation method such as determining the distribution is an example of a calculation method of the ratio of presentation that can use the UPV management table in the example of FIG. Hereinafter, such a method is referred to as a first ratio calculation method in order to distinguish it from other methods.

  Specifically, for example, in the UPV management table in the example of FIG. 2, “period 1” is one session, and the distribution ratio is determined according to the access frequency of red wine and white wine in “period 1”. Suppose that it is decided. In this case, the number of updates for “UPV1” in “period 1” (1 session) for red wine is “3 times”, and “total of each period” for “period 1” is 4 times. The presentation ratio (allocation ratio) will be 3/4. Similarly, the number of updates for “UPV2” during “period 1” (1 session) for white wine is “1 time”, and “total of each period” for “period 1” is “4 times”. Therefore, the ratio of white wine presentation (distribution ratio) is 1/4. Accordingly, if four wines can be presented to the user as recommended wines, three red wines and one white wine are presented to the user as recommended wines.

  Next, an example of a method for calculating the ratio of presentation that can use the UPV management table in the example of FIG. 3 will be described.

  That is, the calculation method of determining the display distribution for each UPV according to the state of access to each page of the user is an example of the calculation method of the ratio of presentation that can use the UPV management table of the example of FIG. is there. In other words, in the UPV management table of the example of FIG. 3, the calculation method of determining the display distribution for each UPV by dividing the “integrated points” of each UPV by the “total” of “integrated points” It is an example of the calculation method of the ratio of presentation which can utilize the UPV management table of the example of FIG. Hereinafter, this method will be referred to as a second ratio calculation method in order to distinguish it from other methods.

  Specifically, for example, in the UPV management table in the example of FIG. 3, “UPV1” “integrated points” for red wine is “44”, and “total” of “integrated points” is “80”. Therefore, the ratio of red wine presentation (distribution ratio) is 44/80 = about 0.6. Similarly, the “accumulation point” of “UPV2” for white wine is “18”, and the “total” of “accumulation points” is “80”, so the ratio of white wine presentation (allocation ratio) Becomes 18/80 = about 0.2. Therefore, if ten wines can be presented to the user as recommended wines, six red wines, two white wines, and two other wines (such as sparkling wine) are presented to the user as recommended wines.

  Further, the UPV management table in the example of FIG. 2 and the UPV management table in the example of FIG. 3 are combined to generate a UPV management table that describes the result of accumulating “accumulation points” for each period, and adopt it. be able to. That is, the UPV management table having the information in the second column and the third column in the example of FIG. 4 described above can be adopted. When such a UPV management table is adopted, it is also possible to adopt a method of determining a recommended ratio suitable for a scene using “accumulated points” accumulated for each period. For example, on the top page of the site, the recommendation rate is determined using the “accumulated points” for the entire period, and the lower-level pages such as the settlement page and basket (shopping basket) content confirmation page are short-term (one session or one day). A method of determining a recommendation ratio using “accumulated points” can also be adopted. Hereinafter, this method will be referred to as a third ratio calculation method in order to distinguish it from other methods.

  Next, an example of a method for calculating the ratio of presentation that can use the UPV management table in the example of FIG. 4 will be described.

  That is, the calculation method of estimating the next operation of the user based on each transition probability and determining the display distribution for each UPV based on the estimated result is a presentation that can use the UPV management table in the example of FIG. It is an example of the calculation method of the ratio. Hereinafter, this method will be referred to as a fourth ratio calculation method in order to distinguish it from other methods.

  Specifically, for example, in the UPV management table in the example of FIG. 4, the “progress probability to UPV3” for sparkling wine is highest from “UPV2” for white wine to “0.45”. In most cases, the purchase operation of sparkling wine is performed after the purchase operation of white wine. Therefore, the display distribution determination unit 43 in FIG. 1 applies this fourth ratio calculation method. For example, when an operation of pressing a purchase determination key for a predetermined white wine is performed, then, for example, a sparkling wine It is possible to change the ratio of the presentation of each type (red wine, white wine, and sparkling wine) so that the distribution ratio of the wine is increased.

  Also, for example, in the UPV management table of the example of FIG. 4, the “transition probability to“ UPV2 ”for white wine from“ UPV3 ”for sparkling wine is also“ 0.32 ”, which is relatively high. After all, when viewed in a predetermined period such as one session, it can often be said that white wine and sparkling wine are often purchased at the same time (many such operations are performed). Therefore, the display distribution determination unit 43 applies the fourth ratio calculation method, for example, when the operation of putting predetermined sparkling wine into a basket (shopping basket) is performed, for example, the distribution ratio of white wine is high. Thus, the ratio of each type of presentation can be changed.

  In the above-described example, the transition probability is used to calculate the presentation ratio. However, the transition probability can also be used to determine the UPV that is the target of the next recommendation process. Specifically, for example, after the operation of pressing the purchase determination key for a predetermined white wine is performed, the UPV selection unit 41 in FIG. 1 can select “UPV3” for sparkling wine, for example. Similarly, for example, when an operation for putting predetermined sparkling wine into a basket (shopping basket) is performed, the UPV selection unit 41 can select “UPV2” indicating white wine.

  In addition, for example, by adopting a UPV management table that accumulates the usage frequency of UPV every month, it is also possible to adopt a method of determining the ratio of presentation so as to produce a change in recommended content according to the season . For example, in December (Christmas), if it can be judged from the UPV management table that the frequency of use of “UPV3” of sparkling wine is high, it seems that more sparkling wine is recommended in the December wine recommendation. Alternatively, it is possible to adopt a method of changing the ratio of presentation of each type (red wine, white wine, and sparkling wine). Hereinafter, this method is referred to as a fifth ratio calculation method in order to distinguish it from other methods.

  Incidentally, the operation history management unit 11 in FIG. 1 manages the user operation history for each property of the item, for example, in the form of UPV in the present embodiment, but the management form is the present embodiment. It is not specifically limited to. For example, the operation history management unit 11 manages the operation history information as it is, that is, stores the operation history information in the UPV / UPV management table storage unit 12 and uses only the necessary operation history information when necessary. Can also be generated. As a result, for example, the operation history management unit 11 can store and manage operation history information separately in the UPV / UPV management table storage unit 12 for short-term and long-term, and as a result, the short-term It is possible to individually generate a UPV based on only operation history information (hereinafter referred to as a short-term UPV) and a UPV based only on long-term operation history information (hereinafter referred to as a long-term UPV). In this case, the recommendation list generation unit 14 can easily determine a recommended item suitable for the current situation by using the short-term UPV and the long-term UPV together.

  Specifically, for example, in the basket (shopping basket) page, the operation history management unit 11 uses only the operation history information of the user during one session, and UPV (for example, red wine, white, etc.) for each property of the item. The recommendation list generation unit 14 can determine each item of each property as a recommended item by using the UPV for each property of the wine and sparkling wine. And the recommendation list production | generation part 14 can determine the ratio of presentation for every property from the access frequency (or situation) to UPV for every property in the session used as object.

  Also, for example, on the top page of the site, the operation history management unit 11 uses all of the user's past operation history information to generate UPVs for each property of the item, and the recommendation list generation unit 14 Using the UPV for each property, each item for each property can be determined as a recommended item. And the recommendation list production | generation part 14 can determine the ratio of presentation for every property from the access frequency (or situation) to UPV for every property over all past periods.

  For example, on the page after purchase, the recommendation list generation unit 14 determines the recommended item again using the UPV generated from all the past operation history information, and the UPV for each property in the target session. The rate of presentation for each property can be determined from the access frequency (or situation). As a result, the recommendation list generation unit 14 can determine, as a recommended item, an item that expands the possibility of additional order, taking into account past purchase details and purchase distribution of the day.

  Heretofore, the functional configuration example of the recommendation server 1 in the recommendation system of FIG. 1 has been described. Next, functional examples of the application server 2 and the user device 3 will be described.

  The application server 2 is configured to include an operation history acquisition unit 21 to a display content generation unit 24.

  The operation history acquisition unit 21 acquires the operation content of the operation unit 31 of the user device 3 via the network 4 and supplies it to the recommendation server 1.

  The service providing unit 22 provides various services to the display content generation unit 24. Specifically, for example, a site for providing (selling) items as described above is an example of a service provided by the service providing unit 22.

  The recommendation acquisition unit 23 supplies the operation content of the operation unit 31 used for the item recommendation process, the item recommendation request, and the like to the recommendation list generation unit 14 of the recommendation server 1. The recommendation acquisition unit 23 acquires the recommendation list generated by the recommendation list generation unit 14 that has received the recommendation request, and supplies the recommendation list to the display content generation unit 24.

  The display content generation unit 24 generates an image (data) as display content of the display unit 32 of the user device 32 and provides the image (data) to the display unit 32 of the user device 3 via the network 4. The image (data) that is the display content of the display unit 32 of the user device 32 refers to, for example, a site image provided from the service providing unit 22 or a recommendation list image supplied from the recommendation obtaining unit 23. .

  The user device 3 is configured to include at least an operation unit 31 and a display unit 32.

  That is, the user performs various operations by operating the operation unit 31 while viewing the display content (item sales site and item recommendation list) displayed on the display unit 32. In addition, about a user's operation, since some specific examples have already been mentioned above, it abbreviate | omits here.

  Next, an example of processing of the recommendation server 1 in the recommendation system of FIG. 1 will be described with reference to the flowchart of FIG.

  In step S1, the operation history management unit 11 of the recommendation server 1 determines whether or not a user operation has been performed.

  If no information is supplied from the operation history acquisition unit 21 of the application server 2, it is determined in step S1 that no user operation is performed, and the process proceeds to step S3.

  On the other hand, when operation history information (operation content) about a predetermined user is supplied from the operation history acquisition unit 21 of the application server 2, the operation history management unit 11 performs a user operation in step S1. In step S2, the UPV / UPV management table of the target user stored in the UPV / UPV management table storage unit 12 is generated or updated. Thereby, a process progresses to step S3.

  As described above, the operation history management unit 11 stores the operation history information supplied from the application server 2 as it is in the UPV / UPV management table without generating or updating the UPV at the time of the process of step S2. It may be stored and managed in the unit 12, and when it is determined that a recommendation request is made in the process of step S3 described later, the operation history information may be used to generate or update the UPV.

  In step S <b> 3, the recommendation list generation unit 14 of the recommendation server 1 determines whether or not a recommendation request has been made from the recommendation acquisition unit 23 of the application server 2.

  If it is determined in step S3 that no recommendation request has been made, the process proceeds to step S5 without performing the process in step S4.

  On the other hand, if it is determined in step S3 that a recommendation request has been made, the process proceeds to step S4. In step S <b> 4, the recommendation list generation unit 14 generates the above-described recommendation list and supplies the recommendation list to the recommendation acquisition unit 23 of the application server 2. Hereinafter, the processing in step S4 is referred to as “recommendation list generation processing”. Details of the “recommendation list generation process” will be described later with reference to the flowchart of FIG.

  When the “recommendation list generation process” in step S4 ends, the process proceeds to step S5. In step S5, the recommendation server 1 determines whether or not an instruction to end the process has been given.

  If it is determined in step S5 that the end of the process has been instructed, the process of the recommendation server 1 ends.

  On the other hand, if it is determined that the end of the process has not been instructed yet, the process returns to step S1, and the subsequent processes are repeatedly executed.

  In the processing example of FIG. 5, if a user operation is performed during the “recommendation list generation processing” in step S 4, the processing in step S 2, that is, the UPV and UPV management table is performed. Is not generated or updated. However, as described above, each of the operation history management unit 11 and the recommendation list generation unit 14 can execute processing independently of each other. That is, in practice, the processes of steps S1 and S2 and the processes of steps S3 and S4 are executed in parallel independently of each other.

  Here, a detailed example of the “recommendation list generation processing” in step S4 will be described with reference to the flowchart of FIG.

  In step S21, the UPV selection unit 41 of the recommendation list generation unit 14 in FIG. 1 selects one or more UPVs of the target user.

  Specifically, in step S21, the UPV selection unit 41 determines the property of the recommendation target by analyzing the recommendation request, and selects the UPV for the property of the recommendation target from the UPV / UPV management table storage unit 12. And is supplied to the recommendation selection unit 42. At this time, the UPV selection unit 41 can naturally select the UPV in consideration of the short-term history information and the long-term history information described above.

  In step S22, the recommendation selection unit 42 determines a recommended item for each selected UPV. That is, in the process of step S22, one or more recommended items are individually determined for each property to be recommended.

  In step S23, the display allocation determination unit 43 refers to the UPV management table of the target user stored in the UPV / UPV management table storage unit 12, and determines the presentation ratio (allocation ratio) for each property of the recommendation target. Decide each.

  In step S24, the recommendation selection unit 42 individually determines the recommended items to be included in the recommendation list for each UPV (for each property of the recommendation target). That is, in the process of step S24, for each property to be recommended, the number of recommended items based on the presentation ratio (distribution ratio) determined in the process of step S23 (this number is often different for each property). Are determined individually.

  In step S25, the recommendation selection unit 42 sorts the recommended items for each property determined in the process of step S24 in order of recommendation level (that is, the recommendation level, here, the similarity level itself). Generate a recommendation list.

  In step S <b> 26, the recommendation selection unit 42 transmits the recommendation list to the recommendation acquisition unit 23 of the application server 2.

  This completes the “recommendation list generation process”. That is, the process of step S4 in FIG. 5 ends, and the process proceeds to step S5.

  Heretofore, an example of processing of the recommendation server 1 has been described.

  By the way, the series of processes described above can be executed by hardware, but can also be executed by software.

  In this case, each of the recommendation server 1, the application server 2, and the user equipment 3 of the recommendation system in FIG. 1 can be configured by, for example, a personal computer shown in FIG.

  In FIG. 7, a CPU (Central Processing Unit) 101 executes various processes according to a program recorded in a ROM (Read Only Memory) 102 or a program loaded from a storage unit 108 to a RAM (Random Access Memory) 103. To do. The RAM 103 also appropriately stores data necessary for the CPU 101 to execute various processes.

  The CPU 101, ROM 102, and RAM 103 are connected to each other via a bus 104. An input / output interface 105 is also connected to the bus 104.

  The input / output interface 105 includes an input unit 106 including a keyboard and a mouse, an output unit 107 including a display, a storage unit 108 including a hard disk, and a communication unit 109 including a modem and a terminal adapter. It is connected. The communication unit 109 controls communication performed with other devices (not shown) via a network including the Internet.

  A drive 110 is also connected to the input / output interface 105 as necessary, and a removable recording medium 111 made of a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is appropriately installed, and a computer program read from them is read. Are installed in the storage unit 108 as necessary.

  When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, a general-purpose personal computer is installed from a network or a recording medium.

  As shown in FIG. 7, the recording medium including such a program is distributed to provide a program to the user separately from the apparatus main body, and a magnetic disk (including a floppy disk) on which the program is recorded. , Removable recording media (packages) consisting of optical disks (including compact disk-read only memory (CD-ROM), DVD (digital versatile disk)), magneto-optical disks (including MD (mini-disk)), or semiconductor memory (Media) 111, but also a ROM 102 on which a program is recorded and a hard disk included in the storage unit 108 provided to the user in a state of being incorporated in the apparatus main body in advance.

  In addition, in the example mentioned above, three types of properties such as red wine, white wine, and sparkling wine were used in the above-described example, but naturally, the property is not limited to this example, and items can be classified into two or more types. Various properties can be used. That is, a UPV having an arbitrary property can be adopted. Specifically, for example, if the item is wine, UPV by price range (for example, UPV of less than 10,000 yen and UPV of 10,000 yen or more, etc.), UPV by grape variety (for example, Cabernet Sauvignon type UPV, Merlot UPV, Pinot Noir UPV, etc., by region (for example, French wine, Bordeaux UPV, Burgundy UPV, Loire UPV, Côte du Rhone UPV, UPV for each region such as Champagne UPV), UPV by producer (for example, for Burgundy wine, it is UPV for each domain or maison, and for California wine etc., it is UPV for each winery) , UPV by age (for example, UPV for old liquor and other UPV), and UPV by sweetness (for example, German wine, Supretese UPV, Auslese UPV, etc.) Prescribed by law It is possible to adopt UPV of various properties such as UPV for each type).

  However, instead of using all of these various UPVs, as described above, it is preferable to use UPVs having properties that take into account differences in user preference trends. In other words, it is preferable to use UPVs having different properties so that there is a clear difference (difference) between the UPVs of each property. In this case, as a scale for measuring the difference, a statistical advantage difference, information engineering information amount (entropy), or the like can be used. These measures can be obtained, for example, by extracting and calculating data of a plurality of users from data obtained through experiments in advance or data obtained from actual services. Therefore, designers or the like should obtain these measures in advance and adopt UPVs of various properties that can be expected to have a clear difference.

  Furthermore, it is possible to hierarchize predetermined properties, generate UPVs for each type belonging to each layer, and use those UPVs. Specifically, for example, in the case of UPV by production area, in the case of French wine, AOC (Appellation d'Origine) defined by the AC (Appellation Controlee) law (ie, the Origin Control Name Law), which is the French wine law. Controlee), ie, the local level is defined as the first layer, the district level is defined as the second layer, and the village level is defined as the third layer. Each of the first to third layers It is possible to generate and use multiple types of UPVs.

  That is, for example, as the first-layer UPV, the above-mentioned Bordeaux UPV can be generated and used. Also, for example, the second tier UPV for the Bordeaux region includes the UPV in the Medoc region, the UPV in the Grave (Pessac Leo Nyan) region, the UPV in the Saint-Emilion region, the UPV in the Pomroll region, the UPV in the Sauternes region, and other UPVs. Can be generated and used. Furthermore, for example, the third-tier UPV for the Medoc region is to generate and use each of the UPV of Saint-Estef village, the UPV of Pauillac village, the UPV of Saint Julian village, the UPV of Margaux village, and other UPVs It is possible.

  In addition, in the case of hierarchization, it is preferable to use UPV up to a hierarchy where there is a clear difference (difference) between UPVs for each property.

  In the present specification, the step of describing the program recorded on the recording medium is not limited to the processing performed in time series along the order, but is not necessarily performed in time series, either in parallel or individually. The process to be executed is also included.

  Further, in the present specification, the system represents the entire apparatus including a plurality of apparatuses and processing units.

It is a block diagram which shows the functional structural example of the information processing system or apparatus to which this invention is applied. It is a figure which shows an example of a UPV management table. It is a figure which shows an example of a UPV management table. It is a figure which shows an example of a UPV management table. It is a flowchart explaining the example of a process of the server of FIG. It is a flowchart explaining the detailed example of the "recommendation request production | generation process" of step S4 of FIG. It is a figure which shows the hardware structural example of the information processing apparatus of FIG.

Explanation of symbols

  1 recommendation server, 2 application server, 3 user equipment, 4 network, 11 operation history management unit, 12 UPV / UPV management table storage unit, 13 item feature vector storage unit, 14 recommendation list generation unit, 41 UPV selection unit, 42 recommendation Selection unit, 43 display allocation determination unit, 101 CPU, 102 ROM, 103 RAM, 108 storage unit, 111 removable recording medium

Claims (6)

For each of a plurality of properties that an item can take, preference vector generation that individually generates a user preference vector representing the user's preference using a user's past operation history for each of one or more items having the corresponding property Means,
A property of a recommendation target is determined from among the plurality of properties, and a user preference vector for the property of the recommendation target among the user preference vectors for each of the plurality of properties generated by the preference vector generation unit; , Calculating a similarity with a feature vector indicating a feature of each of the one or more items having the property to be recommended, and calculating the item corresponding to the feature vector satisfying a predetermined condition with the calculated similarity An information processing apparatus comprising: a recommendation unit that determines a recommended item to be recommended to the computer. And further comprising storage means for storing the operation history of the user.
The information processing according to claim 1, wherein the preference vector generation unit further performs control to store the operation history of the user in the storage unit in a form that can be managed individually for each of the plurality of properties. apparatus. The recommendation unit determines two or more of the plurality of properties as the properties of the recommendation target, and individually determines one or more recommended items for each of the properties of the two or more recommendation targets. The information processing apparatus according to claim 2. When at least a part of the recommended items determined for each property of the recommendation target by the recommendation unit is presented to the user, the recommendation target is obtained using the operation history stored in the storage unit. The information processing apparatus according to claim 3, further comprising: a distribution determination unit that determines a distribution ratio of presentation for each of the properties. For each of a plurality of properties that an item can take, preference vector generation that individually generates a user preference vector representing the user's preference using a user's past operation history for each of one or more items having the corresponding property Steps,
The user preference for the property of the recommendation target among the user preference vectors for each of the plurality of properties generated by determining the property of the recommendation target from the plurality of properties and processing the preference vector generation step. A similarity between a vector and a feature vector indicating each feature of one or more items having the property to be recommended is calculated, and an item corresponding to the feature vector in which the calculated similarity satisfies a predetermined condition, A recommendation step of determining as a recommended item to be recommended to the user. A program for causing a computer that controls a recommendation device to recommend an item to a user,
A preference vector that individually generates a user preference vector representing the user's preference by using the user's past operation history for each of one or more items having the corresponding property for each of a plurality of properties that the item can take Generation step;
The user preference for the property of the recommendation target among the user preference vectors for each of the plurality of properties generated by determining the property of the recommendation target from the plurality of properties and processing the preference vector generation step. A similarity between a vector and a feature vector indicating each feature of one or more items having the property to be recommended is calculated, and an item corresponding to the feature vector in which the calculated similarity satisfies a predetermined condition, A recommendation step of determining as a recommended item to be recommended to the user.

Images