JP2009223844A - Related information search apparatus, related information search method, related information search program, and recording medium with this program recorded - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a search technique for implementing search reflecting users' preferences. <P>SOLUTION: A related information search apparatus 100 comprises a history data processing part 130 for generating, from target users' history data, user-specific session data represented by user ID, item ID and access time and computing, from the session data, the number of sessions including a given item, the number of sessions including an ordered item pair and the number of sessions including two given items as a pair, a computational processing part 140 for calculating parameters using the session numbers as indices and computing a degree of association between items using the indices, an input processing part 120 for receiving a search condition, a scoring part 150 for generating a ranking list including related items related to an item indicated by the search condition according to the degree of association computed about the item indicated by the search condition, and an output processing part 160 for outputting the ranking list. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a search technique, and in particular, searches for related objects related to an object of a search condition using an analysis result of history data indicating an action history of a user performing an action of accessing an object such as an item. Related to search technology.

  Conventionally, a mechanism for identifying a user (browser or buyer) in an online store on the Internet and changing the services and contents provided to the user in accordance with the behavior of the user accessing an object such as an item, so-called personality Lizaization is being used. For example, a method for personalizing a ranking of search results based on a relationship between a user and another user (a relationship between users) and an action history of the user accessing an item has been proposed (Non-Patent Document 1). reference).

  The method described in Non-Patent Document 1 analyzes each event (event) that a user accesses an item according to a time series. For example, when user A continues to access item 1 and item 2, and user B continues to access item 1 and item 2 at a later time, user B's action is user A's action. It is thought that it is following. Or it is thought that the user A is affecting the action of the user B. In this case, a user's action can be predicted using information of a user having a similar action history. According to the technology described in Non-Patent Document 1, music or video purchased by each user can be predicted using the purchase history of the user in an online music download and video viewing service.

  Further, in the prior art, in order to define the relevance of items, text data (hereinafter simply referred to as text) included in the items, metadata thereof, and the like are used. As a result, the relevance of items defined by the conventional technology is, for example, the relationship between two items by using the criteria of whether the item belongs to a category or genre such as “economy” or “science”. Can be defined as an explicit relationship.

Further, in the conventional technology, when analyzing a user's history, information indicating the relationship between items is often extracted based on the “number of co-occurring users”. For example, if the users who access item 1 are “user A” and “user B”, “number of co-occurring users” for item 1 is “2”, and similarly, “co-occurrence” for each item. If the “number of users” is calculated, items similar in “number of co-occurring users” are considered to be closely related.
Noriaki Kawamae, Takeshi Yamada, Nobuo Ueda, "Proposal of Personalization Method by Discovery of Relative Innovator", Information Science and Technology Letters, 2007, vol.6, pp.99-102

  In a search using a user's action history to obtain an item (object) to be recommended (recommended) to the user, a search is performed when a related item is output as a search result for an item input as a search condition. It is desired that various searches can be made flexibly corresponding to the purpose of the user. For example, there is a search need that reflects subjective criteria (user preferences) such as “relax” and “interesting”. For example, whether or not one movie as an item feels “interesting” varies depending on the user, but among users who form groups with similar hobbies and preferences, the favorite movie that feels “interesting” is similar It is done. In this case, “movie B” purchased by a user in a group with similar hobbies and preferences has not been seen yet, but “movie B” has been purchased. "Movie A" is recommended as a movie related to "

  However, in the conventional technology, the relevance of an item is defined by the text included in the item, its metadata, etc., and the definition of the relevance of the item is different. Is not applicable. In other words, in the conventional technology, the relationship between items can be defined as an explicit relationship such as “economy” or “science”, but an implicit relationship such as “relaxation” or “interesting” that reflects user preferences. Cannot be defined. Therefore, for example, even if the user tries to search for an item to be recommended as a movie that is “relaxed”, the intended search result cannot be obtained. Furthermore, in the conventional technique, if there is no text or metadata associated with an item, it becomes difficult to calculate the relevance.

  In addition, in the conventional technology, when analyzing a user's history, information on the relationship between items is often extracted based on the “number of co-occurring users”, and the relationship between items is analyzed in time series. Until that time, it was not considered enough. For this reason, the related item as the search result does not reflect the change of the “number of users accessing the item” over time. Accordingly, there is a demand for a search that reflects the user's preferences while considering the time-series change when extracting the information related to the items.

  Therefore, the present invention has been made in view of the above problems, and provides a search technique capable of performing a search reflecting the user's preference while taking into account the time series change of the relationship between objects. For the purpose.

  In order to solve the above-mentioned problem, the related information search device according to the present invention provides an object index indicating an object and a relationship between objects obtained by analyzing in advance history data indicating an action history of a user performing an action of accessing an object. Based on the above, a search condition is used as at least one of the objects to be analyzed, and a related object related to the search condition is searched as related information from other objects to be analyzed. An information search device, which is represented by a set of user identification information, object identification information, and access time information indicating a time at which the user performs an action to access the object from the target history data. Individual session data for each user is generated, and a predetermined one is generated from the generated session data. The number of sessions in which an object has appeared, the number of sessions in which two predetermined objects whose access order is specified in the object pair have appeared, and the number of sessions in which the predetermined object is included as one of the object pairs are used as the statistics of the history data A history data processing means to be obtained and a parameter using the number of sessions obtained as a statistic of the history data are calculated as the index, and the degree of association between any two objects based on the calculated index Calculation processing means for calculating each of the above, an input processing means for receiving the search condition, and another object that is the object of the analysis based on the relevance calculated in advance with respect to the object indicated by the received search condition Of these, the options indicated in the search conditions It characterized in that it comprises a scoring means for generating a ranking list including at least the most relevant related objects, and an output processing means for outputting the ranking list to the object.

  In addition, in order to solve the above-described problem, the related information search method according to the present invention provides a relationship between an object and an object obtained by analyzing in advance history data indicating an action history of a user performing an action of accessing an object. Based on the index indicated, at least one of the objects to be analyzed is used as a search condition, and the related object related to the search condition is searched as related information from among other objects to be analyzed. A related information search method for a related information search device, wherein the related information search device includes an input processing unit, a history data processing unit, a calculation processing unit, a scoring unit, and an output processing unit, The history data processing means accepts the target history data, and from the received history data, user identification information and From the step of generating individual session data for each user represented by a set of object identification information and access time information indicating a time at which the user performs an action to access the object, from the generated session data, The history data includes the number of sessions in which a predetermined object has appeared, the number of sessions in which a predetermined two objects whose access order is specified in the object pair have appeared, and the number of sessions in which the predetermined object is included as one of the object pairs. And calculating the parameter using the number of sessions determined as the statistic of the history data as the index, and based on the calculated index Any two objects Each step of calculating the degree of association between the input processing unit, the step of receiving the search condition, and the scoring unit calculated in advance for the object indicated by the received search condition Executing the step of generating a ranking list including at least the related objects most relevant to the object indicated by the search condition among the other objects to be analyzed based on the degree of association, and the output process The means executes the step of outputting the ranking list.

  According to the related information search device having such a configuration or the related information search method of such a procedure, the related information search device is configured to perform user identification information and object identification information from target history data in advance before executing the search. Then, session data including access time information is generated for each user, and from the session data, for example, a session in which two predetermined objects with the access order specified in the object pair appear as a history data statistic. Then, the related information search device calculates, as an index, a parameter using the number of sessions obtained as a statistic of the history data, and based on the calculated index, each degree of association between any two objects is calculated in advance. Calculate it. Here, the object includes, for example, computerized information, real-world information, content, and items, and the access includes, for example, operation information such as login, browsing, download, and purchase. The access condition indicates, for example, a reference range, a reference direction, an access time interval threshold value, and an access order interval threshold value of a target object pair. According to this, unlike the conventional case, the related information search device obtains the number of sessions including objects and object pairs and the degree of association between objects based on the number of sessions, thereby identifying user identification information, object identification information and The access time information can be analyzed according to a time series.

  When the related information search device executes the search, based on the relevance calculated in advance with respect to the object indicated by the received search condition, the related information search device applies to the object indicated by the search condition. A ranking list including at least the most related objects is generated and output. This ranking list is generated based on the degree of association between objects defined based on the number of sessions. Therefore, since the degree of association between objects defined based on the number of sessions does not use text or its metadata, the relationship between two items can be defined as an implicit relationship. Therefore, the related information search device enables a search reflecting the user's preference while taking into account the time series change of the relationship between objects. Thereby, the related object reflecting a user preference can be recommended to the user.

  Further, in the related information search device according to the present invention, the history data processing means accesses the time of access to any two objects from the number of sessions determined as the statistics of the history data and the individual session data. A parameter indicating time average based on the average value of the time obtained as the time statistic. It is preferable to include a time average calculating unit that calculates the degree of association as an index, and a second relevance degree calculating unit that calculates the degree of association between any two objects based on the parameter indicating the calculated time average. .

  Further, in the related information search method according to the present invention, the history data processing unit includes a time statistic calculation unit, and the time statistic calculation unit includes the number of sessions obtained as a statistic of the history data, A step of obtaining an average value of a time indicating a difference in access time to any two objects from individual session data as a time statistic, and the calculation processing means includes a time average calculation means and a second relevance degree Calculation means, and the time average calculation means, based on the average value of the time obtained as the time statistic, executing a step of calculating a parameter indicating the time average as an index at the time of the search, The second relevance calculating means executes a step of calculating relevance between any two objects based on the calculated parameter indicating the time average. It is preferable to.

  According to the related information search device having such a configuration, or the related information search method of such a procedure, the related information search device can perform arbitrary processing from the session data generated in advance and the number of sessions determined in advance before executing the search. The average value of the time indicating the difference between the access times of the two objects is obtained as a time statistic, the parameter indicating the time average is calculated as an index, and the relevance is calculated based on the parameter indicating the time average. . Therefore, when executing the search, the related information search device generates a ranking list based on the relevance calculated in advance based on the parameter indicating the time average for the object indicated by the received search condition. Can do. According to this, in addition to the number of sessions, the related information search device can take into account the time series change of the relationship between objects by obtaining the average of the differences in the access times to any two objects. In addition, the parameter indicating the time average and the number of sessions can be used to predict the cumulative number of users accessing the item and the cumulative number of accessed users at each time as the number of users accessing the item.

  Further, in the related information search device according to the present invention, the history data processing means allows the user to move one of the object pairs from the number of sessions determined as the statistics of the history data and the individual session data. An object number statistic calculating unit that obtains an average value of the number of objects indicating the number of different objects accessed from the access to the other as an object number statistic, and the calculation processing unit includes the object number statistic Based on the average value of the number of objects determined as a quantity, based on the parameter indicating the average of the number of different objects, the different object number calculation means for calculating a parameter indicating the average of the number of different objects as the index The degree of association between any two objects It is preferable to provide a third degree of association calculation means for calculating, respectively.

  Further, in the related information search method according to the present invention, the history data processing unit includes an object number statistic calculating unit, and the object number statistic calculating unit includes the number of sessions obtained as a statistic of the history data, and And obtaining an average value of the number of objects indicating the number of different objects accessed from when the user accesses one of the object pairs to the other from the individual session data as an object number statistic. The calculation processing means includes a different object number calculation means and a third relevance calculation means, and the different object number calculation means sets the average value of the number of objects obtained as the object number statistic. Based on the parameter indicating the average of the number of different objects. A step of calculating as a time index, wherein the third relevance calculating means calculates a relevance between any two objects based on the calculated parameter indicating the average number of different objects. Is preferably performed.

  According to the related information search device having the above configuration or the related information search method of the procedure, the related information search device uses the object pair from the session data generated in advance and the number of sessions determined in advance before executing the search. The average value of the number of different objects accessed before the transition between them is calculated as an object number statistic, and the parameter indicating the average number of different objects is calculated as an index. The relevance is calculated based on a parameter indicating the average of each. Therefore, when executing the search, the related information search device searches the ranking list based on the relevance calculated in advance based on the parameter indicating the average of the number of different objects for the object indicated by the received search condition. Can be generated. According to this, in addition to the number of sessions, the related information search device can take into account the time series change of the relationship between objects by obtaining the average value of the number of objects indicating the number of different objects described above. . In addition, the number of users accessing an item can be predicted by the parameter indicating the average of the number of different objects and the number of sessions, and the cumulative number of users accessing by an arbitrary time point and the cumulative number of accessing users at each time can be predicted.

  A related information search program according to the present invention causes a computer to realize the function of the related information search apparatus described above. By being configured in this way, a computer in which this program is installed can realize each function based on this program.

  A computer-readable recording medium according to the present invention is characterized in that the related information retrieval program is recorded. By being configured in this way, a computer equipped with this recording medium can realize each function based on a program recorded on this recording medium.

  According to the present invention, the related information search device can perform a search that reflects the user's preference, and can therefore recommend a related object that reflects the user's preference to the user. In addition, the related information search device enables a search that reflects the user's preference while taking into consideration the time series change of the relationship between objects, so that personalization based on the time series change of the degree of association between objects is possible. To do.

  The best mode (hereinafter referred to as “embodiment”) for carrying out the related information search device and search processing method of the present invention will be described below in detail with reference to the drawings.

[Overview of configuration of related information search device]
FIG. 1 is a configuration diagram schematically showing a related information search device according to an embodiment of the present invention.
The related information search device 100 analyzes in advance history data indicating an action history of a user performing an action to access an object for each user, and obtains an index at the time of search indicating an object and a relationship between objects. Further, the related information search apparatus 100 relates to the search condition from among other objects to be analyzed using at least one of the objects to be analyzed as a search condition based on the obtained index. Search related objects as related information. Here, the object includes, for example, computerized information, real-world information, content, and items, and the access includes, for example, operation information such as login, browsing, download, and purchase. Below, an object is demonstrated as an item. In the present embodiment, the related information search device 100 will be described as being included in a search service system such as a search engine or an EC site (a site for performing electronic commerce).

  The related information search apparatus 100 is stored in a RAM (Hard Disk Drive) and a computer such as a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and an input / output interface, for example. Has a program to be deployed. That is, these hardware resources are controlled by a program through the cooperation of the hardware device and software, and the related information search device 100 includes a data storage unit 110 and an input processing unit 120 as shown in FIG. And a history data processing unit 130, a calculation processing unit 140, a scoring unit 150, and an output processing unit 160.

  The data storage unit 110 stores user history data input from the outside, and parameters and relevance (relevance) calculated as calculation processing results. The data storage unit 110 includes, for example, a general hard disk device And stores various programs and various data used by the CPU and the like.

  The input processing unit 120 includes a predetermined input interface, receives various information (data and commands) input from an input device such as a keyboard, a mouse, and a disk drive device (not shown), and receives a history data processing unit 130 and a scoring unit. 150 is input. The input processing unit 120 receives the history data of each user and inputs it to the history data processing unit 130. The user history data indicates a history of user actions such as each user operation such as a user search query, access, and purchase. The input processing unit 120 receives a search condition such as an item ID and inputs it to the scoring unit 150. The input processing unit 120 may be configured by a communication interface that inputs various types of information from a communication network (not shown).

  The history data processing unit (history data processing means) 130 analyzes the user's history data in order to calculate a parameter serving as an index based on the user's history data in the subsequent calculation processing unit 140, The statistics are calculated. Details of the statistics of historical data will be described later.

  The calculation processing unit (calculation processing means) 140 calculates a parameter using the statistical amount of history data as an index, and calculates the degree of association between any two objects based on the calculated index. . The calculation processing unit 140 calculates the relationship between items in advance by calculating the degree of association. Details of parameters indicating indices and relevance will be described later.

  The scoring unit (scoring means) 150 is based on the relevance calculated in advance with respect to the item indicated by the search condition received by the input processing unit 120, with respect to the item indicated by the search condition among other items. Generate a ranking list that includes at least the most relevant items. Here, the ranking list indicates a list of IDs of related items related to the item indicated by the input item ID. In addition, the items indicated by the search conditions and the other items are the objects of analysis by the history data processing unit 130 in advance. In the present embodiment, the scoring unit 150 predicts the number of users accessing the item, as will be described later.

  The output processing unit 160 is constituted by an output interface such as a graphic board, for example, and outputs the calculation processing result by the scoring unit 150 to an output device such as a liquid crystal display (not shown). Here, the calculation processing result is, for example, a ranking list created as a search result.

[Overview of related information search device processing]
Next, an overview of the processing of the related information search device will be described with reference to FIGS. FIG. 2 is an explanatory diagram showing an overview of history data processing in the related information search device shown in FIG. 1, and FIG. 3 is an explanatory diagram showing an overview of search processing in the related information search device shown in FIG. is there. 4 is an explanatory diagram showing an overview of history data of a plurality of users, and FIG. 5 is an explanatory diagram showing an overview of the operation of the related information search device shown in FIG.

  In the present embodiment, the related information search device 100 calculates the relationship between items that appear in the history in advance from the user's history data, as shown by the broken line in FIG. 5, and as shown by the solid line in FIG. When there is a search request of a certain user, a scoring process for calculating related items is executed for the request and the search result.

<Outline of processing to be performed in advance before search>
As illustrated in FIG. 2, the history data processing unit 130 receives user history data 201 via the input processing unit 120. The history data processing unit 130 generates history data statistics 202 based on the user history data 201 and outputs the statistics data 202 to the calculation processing unit 140. Further, the history data processing unit 130 records the statistics 202 of the generated history data in the data storage unit 110. The calculation processing unit 140 generates a parameter for each item and a parameter 203 for each item pair based on the statistics 202 of the history data acquired from the history data processing unit 130, and generates the generated parameter for each item and each item pair. The parameter 203 is recorded in the data storage unit 110. The parameter 203 includes the index and the degree of association between items. Note that the calculation processing unit 140 can also directly output the generated parameter to the scoring unit 150.

<Outline of processing at the time of search request>
As illustrated in FIG. 3, the input processing unit 120 receives an item ID 301 from a search service or the like and passes it to the scoring unit 150. The scoring unit 150 receives the item ID 301 via the input processing unit 120. The scoring unit 150 uses the item ID 301 as a key to call a relevance level between items as a parameter for each item and a parameter 203 for each item pair stored in advance in the data storage unit 110. Next, the scoring unit 150 rearranges the opponents in the order most related to the item ID 301 as the scoring result 302 for the item ID input with the data (relationship between items) called from the data storage unit 110. The ranking result 303 is generated and output to the output processing unit 160. Note that an item name may be accepted instead of the item ID 301. Moreover, you may make it input user ID with item ID.

<Search example>
After analyzing the user history data in advance in the related information search apparatus 100 with the content “movie” and the item “movie title”, for example, “item” indicating “Leon” as the movie title is analyzed. When “ID” is input, a ranking list of “item ID” indicating “Jackal”, “Fifth Element”, and the like is output as a movie title. These movies are not related in the category or metadata by the conventional search device. However, it is a popular movie for users with the same preference. For example, after “Leon” is released, “Jackal”, “Fifth Element”, and the like are released. `` Leon '' and `` Fifth Element '' are the same producers and performers, `` Jackal '' and `` Fifth Element '' are the same performers, `` Leon '' and `` Jackal '' are great killer Is an important cast. Therefore, it is possible to perform a search reflecting the user's preference that cannot be defined only by the text included in the item and its metadata as in the conventional case.

<Target history data>
Here, an overview of history data to be handled by the history data processing unit 130 will be described with reference to FIG. In this example, four users (user A, user B, user C, user D) are assumed. Details of the history data 201a, 201b, 201c, and 201d of the four users are described in the history data 201 of all users as target history data. The history data processing unit 130 collects history data 201 for all users. The history data of each user has at least a user ID, an item ID, and an access time. For example, it is understood that the user A accesses “item i ₁ ” on January 1 (Jan. 1) and accesses “item i ₂ ” on January 31 (Jan. 31). . In addition, you may add the kind (login, purchase, browsing, download) of a user's action regarding operation which accesses an item as information contained in historical data.

The history data processing unit 130 can calculate a statistic between items via the user as the history data statistic 202 from the history data 201 of all users. In FIG. 4, the time required for each user to transition from “item i ₁ ” to “item i ₂ ” is illustrated by an arrow starting from item 401. For example, the user A requires 30 days from the date when “item i ₁ ” is accessed to the day when “item i ₂ ” is accessed. Similarly, the time required for each user to change from “item i ₁ ” to “item i ₃ ” and the time required for each user to change from “item i ₁ ” to “item i ₄ ” It is shown. However, since the reference direction condition (access condition) is “backward”, the user D who accesses “item i ₂ ” before “item i ₁ ” is not illustrated.

[Detailed configuration of related information search device]
FIG. 6 is a block diagram illustrating an example of a detailed configuration of the related information search apparatus illustrated in FIG.
(History data processing part)
Here, the history data processing unit 130 extracts the history data statistics 202 (see FIG. 2), as shown in FIG. 6, the session data generation means 131, the session statistics calculation means 132, and the time The statistic calculation unit 133 and the item count statistic calculation unit 134 are provided.

<Session data generation means>
The session data generating unit 131 uses a user ID (user identification information), an item ID (item identification information), and an access time (access time) indicating the time when the user performs an action to access the item from the target history data. Each session data (hereinafter, simply referred to as a session) for each user represented by a pair of information) is generated.

  The session data generating unit 131 obtains data (hereinafter referred to as a sequence) obtained by dividing the target history data into user units. Here, the target history data is recorded by a server or the like. The sequence has a unique ID, and has both an item ID and an access time inside the sequence. The history data processing unit 130 can divide the sequence based on conditions set by the user. In the present embodiment, when duplicate items appear during sequence generation, the session data generating unit 131 uses only the first appearing item and its access time. The reason is that when one user accesses the same item again, the experience gained in the first access is reflected, so it is difficult to handle the same as the first access, and the process for that is complicated It is.

The session data generating unit 131 prepares for obtaining a statistic by dividing each sequence into sessions for each item. In the present embodiment, the session indicates a list generated in the format of <item: time when the item is accessed> for each user from the history data of the user. For example, the set u _i of the user's session whose user ID is “i” is i _j (item ID “j”) for an item accessed by the user, and t is the time when the user accessed “item i _j ”. _{Assuming that} there are a total of h sessions as _j , it can be expressed as in equation (1).

u _i = {<i ₁ : t ₁ >, <i ₂ : t ₂ >,..., <i _h : t _h >} Expression (1)

In the following processing, when the same item appears in the session of the same user, only the item that appears first and its time are left. That is, when the user with the user ID “i” accesses the same item _ij repeatedly, only the information of the item at the access time t _j accessed for the first time is left. If necessary, a threshold (time threshold) can be set for the access time interval at which the item is accessed, and the session can be divided when the threshold is exceeded.

<Session statistics calculation means>
The session statistic calculation unit 132 specifies the number of sessions in which a predetermined item appears and the access order in an item pair as the number of sessions that co-occur between items from the session generated by the session data generation unit 131. The number of sessions in which two predetermined items appear and the number of sessions including a predetermined object as one of the object pairs are obtained as statistics of the history data. Here, the item pair is a combination of arbitrary items. An access condition (hereinafter referred to as a reference condition) is a condition that is set when obtaining the number of sessions (counting the number of sessions), such as a reference range, a reference direction, a time threshold, or an item count threshold. It is. When the reference condition is set, only sessions that satisfy the condition in the item pair are counted. In the following, three types of sessions “n _a , n _aj , n _{· j} ”, which differ depending on the counting method, are shown as examples.

(1) Number of sessions “n _a ” in which one predetermined item appears
The number of sessions in which a predetermined item has appeared is the number of sessions including the predetermined item. The number of sessions including item i _a (item ID = a) is “n _a ”. The session a set of one user to be counted, even more sessions including item i _a, only counted once for the user.

(2) Number of sessions “n _aj ” in which two predetermined items whose access order is specified in the item pair appear
The number of sessions in which two predetermined items whose access order is specified in the item pair appears is the number of sessions including the item pair based on a predetermined access condition. There is a reference condition that specifies item ID = a as a starting point, and the number of sessions including a pair (item pair) consisting of item i _a (item ID = a) and item i _j (item ID ≠ a) is “n” _aj ". The session a set of one user to be counted, when the item pairs item i _a and item i _j there is a plurality, in recent sessions for that user, be counted only once.

(3) Number of sessions “n _{· j} ” including a predetermined object as one of the object pairs
The number of sessions including a predetermined object as one of the object pairs is the number of sessions including a predetermined item with each item as a starting point or an ending point. In this case, there is no reference condition for designating a specific item ID = a as a starting point or an ending point. In this case, the number of sessions in which item i _j (item identifier j ≠ ·) appears corresponding to each item i _· (item ID = ·) is assumed to be “n _{· j} ”.

<Time statistics calculation means>
The time statistic calculation means 133 includes two items (item pairs) from the number of sessions obtained as the statistics of the history data by the session statistic calculation means 132 and the individual sessions generated by the session data generation means 131. The average value (time average) of the time indicating the difference in access time is obtained as a time statistic. In this embodiment, since there are two types of sessions (n _aj , n _{· j} ) for the item pair, the time statistic calculation means 133 determines two types of “time average” correspondingly.

として、式（２）により求める。 ≪Time average〜1≫
The time statistic calculation means 133 calculates the time average between items until the transition to “item i _j ” with respect to “item i _a ”.

As described above, it is obtained by the equation (2).

Here, n _aj is the number of sessions calculated by the session statistic calculation unit 132, and indicates the number of sessions in which “item i _j ” appears, satisfying the reference condition from “item i _a ”. S _aj indicates a set of sessions that satisfy the reference condition from “item i _a ” and in which “item i _j ” appears. t _{aj, k} indicates statistical information obtained for each session. k indicates a session ID. However, the number indicated by _naj must be the same as the number of elements of _Saj (the number identified by the session number k). Therefore, strictly speaking, even if there are actually multiple sessions in which a single user accesses the same combination of “item i _a ” and “item i _j ”, The number of sessions of the same combination of “item i _a ” and “item i _j ” is counted only once.

The time statistic calculating unit 133 obtains, for each session _{, a} time interval t _{aj, k} that satisfies a preliminarily designated reference condition (reference range, reference direction, time threshold). Here, the reference range is a range from “item to be a starting point” to “item to be referred to”. For example, if the number of different items “2” is specified as the reference range, the time interval t _{aj, k} is extracted only for items with the number of different items within “2” counting from the “starting item”. To do. Specifically, in the present embodiment, for the user B shown in FIG. 4, for example, “item I ₄ ” and “item I ₃ ” are changed before “item I ₁ ” is changed to “item I ₂ ”. Then, since “item I ₂ ” is reached, the number of different items is defined as “3”. In this case, it is also possible to define the number of different items as “2” as the number of items passed in between. Further, when the “item I ₁ ” that is the same as the starting point is passed along the way, the count can be left or removed. Here, when removed, “different number of items” different from the physical time information can be defined.

In the reference condition, the reference direction is a time-series direction as viewed from the “starting item”. For example, when “both” is specified as the reference direction, the time series viewed from “starting item” does not matter, but when “forward (or backward)” is specified, “starting point” The time interval t _{aj, k} is extracted only for items that are before (or after) the time as viewed from the item. In addition, when a time threshold is set as a reference condition, it is possible not to extract a time interval t _{aj, k} exceeding the time threshold.

を、前記した式（２）により計算すると、式（３）のように求められる。なお、式（３）の計算では小数第２位を四捨五入した。ここで、「アイテムｉ₁」の前に「アイテムｉ₂」を参照しているユーザＤは、参照方向の条件「後」を満たさないので、この計算には使われない（ＮＧ）。 For example, if there is a user of the history data as shown in FIG. 7 (a), as an example of a reference condition, i ₁ the "item as a starting point", see the direction condition as a "rear", "Item i ₁ ”To“ Item i ₂ ”average time between items

Is calculated by the above-described equation (2), it is obtained as in equation (3). In the calculation of equation (3), the second decimal place was rounded off. Here, the user D referring to the “item i ₂ ” before the “item i ₁ ” does not satisfy the condition “after” in the reference direction, and thus is not used for this calculation (NG).

として、式（４）により求める。 ≪Time average〜2≫
In the time average to the first, the time average is calculated between items, but here, the points are calculated for each item. The time statistic calculation means 133 calculates the time average until “item i _j ” appears for each item as a starting point.

As described above, it is obtained by the equation (4).

Here, n _{· j} is the number of sessions calculated by the session statistic calculation unit 132, and indicates the number of sessions in which “item i _j ” appears as seen from all items. S _{· j} represents a set of sessions including items in which “item i _j ” appears later. t _{aj, k} indicates statistical information obtained for each session. k indicates a session ID.

として前記した式（４）により計算すると、式（５）のように求められる。ここでは、「アイテムｉ₁」の前に「アイテムｉ₂」を参照しているユーザＤも計算に使う（ＯＫ）。 For example, in the case where there is user history data as shown in FIG. 7B, as an example, the “time average” until “item i ₂ ” appears is set for all “starting items”. Average time per item

As calculated by the above-described equation (4), it is obtained as in equation (5). Here, the user D who refers to “item i ₂ ” before “item i ₁ ” is also used for the calculation (OK).

<Item statistic calculation means>
The item number statistic calculating unit (object number statistic calculating unit) 134 includes the number of sessions obtained as the statistical data statistic by the session statistic calculating unit 132, the individual sessions generated by the session data generating unit 131, and From the two items (item pairs), the average value of the number of items indicating the number of different items accessed from when the user accesses one to the other (average of the number of different items between items) This is obtained as an item count statistic. In the present embodiment, since there are two types of sessions (n _aj , n _{· j} ) for the item pair, the item number statistic calculation means 134 correspondingly calculates two types of “average of the number of different items”. Ask.

として、式（６）により求める。 ≪Average of the number of different items ~ 1≫
The item count statistic calculation means 134 calculates the average of the number of different items until the transition from “item i _a ” to “item i _j ”, and the average of the number of different items between items.

As described above, it is obtained by the equation (6).

Here, n _aj is the number of sessions calculated by the session statistic calculation unit 132, and indicates the number of sessions in which “item i _j ” appears after “item i _a ”. S _aj indicates a set of sessions in which “item i _j ” appears after “item i _a ”. d _{aj, k} indicates statistical information obtained for each session. k indicates a session ID.

The item count statistic calculation unit 134 determines, for each session _, an interval d _{aj, k} for the number of items that satisfies a reference condition (reference range, reference direction, item count threshold) specified in advance. For example, if an “item number threshold” is set, it is possible not to extract an “item number” exceeding that.

を、前記した式（６）により計算すると、式（７）のように求められる。ここで、「アイテムｉ₁」の前に「アイテムｉ₂」を参照しているユーザＤは、参照方向の条件「後」を満たさないので、この計算には使われない（ＮＧ）。 For example, when there is user history data as shown in FIG. 8A, “item i ₁ ” to “item i ₂ ” are set with “starting item” as i ₁ and reference direction condition as “after”. The average number of items that differ between items up to

Is calculated by the above-described equation (6), it is obtained as in equation (7). Here, the user D referring to the “item i ₂ ” before the “item i ₁ ” does not satisfy the condition “after” in the reference direction, and thus is not used for this calculation (NG).

として、式（８）により求める。 ≪Average of the number of different items ~ 2≫
In the average of the number of different items to the first item, the average of the number of different items is calculated for each item, whereas here, the difference is calculated for each item. The item count statistic calculating means 134 calculates the average of the number of different items until “item i _j ” appears, and the average of the number of different items for each item for all the starting items.

As described above, it is obtained by the equation (8).

Here, n _{· j} is the number of sessions calculated by the session statistic calculation unit 132, and indicates the number of sessions in which “item i _j ” appears as seen from all items. S _{· j} represents a set of sessions including items in which “item i _j ” appears later. d _{aj, k} indicates statistical information obtained for each session. k indicates a session ID.

として前記した式（８）により計算すると、式（９）のように求められる。ここでは、「アイテムｉ₁」の前に「アイテムｉ₂」を参照しているユーザＤも計算に使う（ＯＫ）。 For example, in the case where there is user history data as shown in FIG. 8B, as an example, “average of the number of different items” until “item i ₂ ” appears is “all items starting”. Average number of different items per item for

Is calculated by the above-described equation (8), the equation (9) is obtained. Here, the user D who refers to “item i ₂ ” before “item i ₁ ” is also used for the calculation (OK).

(Calculation processing part)
As shown in FIG. 6, the calculation processing unit (calculation processing unit) 140 includes a session number calculation unit 141, a time average calculation unit 142, a different item number calculation unit 143, a first relevance calculation unit 144, 2 relevance calculating means 145 and third relevance calculating means 146 are provided.

<Session number calculation means>
The number-of-sessions calculating means 141 shows the parameters λ _n, .j shown in Expression (10) and Expression (11) from the number of sessions (n _a , n _aj , n _{· j} ) counted by the session statistic calculating means 132. The parameter λ _{n, aj} is calculated as an index. Note that μ shown in Expression (11) can be set arbitrarily, but can also be determined using a method such as cast sampling. By appropriately setting the value of μ, it is possible to prevent the model from being excessively dependent on the target history data. That is, overfitting with respect to the user's behavior history can be avoided.

<Time average calculation means>
The time average calculation unit 142 calculates a parameter indicating the time average as an index based on the average value of the time (time average between items) obtained as the time statistic by the time statistic calculation unit 133. In this embodiment, the time average calculation unit 142 calculates the equation (12) from the number of sessions ( _naj ) counted by the session statistic calculation unit 132 and the “time average” shown in the above equations (2) and (4). parameter lambda _{t shown),} calculates a parameter lambda _{t, aj} shown in _{· j} and the formula (13). Note that μ shown in Expression (13) can be set arbitrarily, but can also be determined using a method such as cast sampling.

<Different item number calculation means>
The different item number calculating means (different object number calculating means) 143 is different based on the average value of the number of items (average of different items between items) obtained as the item number statistical quantity by the item number statistical quantity calculating means. A parameter indicating an average of the number of items (hereinafter simply referred to as a “parameter of the number of different items”) is calculated as an index. In the present embodiment, the different item number calculating unit 143 determines the number of sessions (n _aj ) counted by the session statistic calculating unit 132 and the “average of the number of different items” shown in the above formulas (6) and (8). The parameter λ _{d, j} shown in the equation (14) and the parameter λ _{d, aj} shown in the equation (15) are calculated. Note that μ shown in Expression (15) can be set arbitrarily, but can also be determined using a method such as cast sampling.

<First relevance calculation means>
The first relevance calculating unit 144 reads the parameter of the number of sessions from the data storage unit 110 and calculates the relevance between any two items among the items to be searched. In this embodiment, it is assumed that the number of sessions that co-occur between items follows a Poisson distribution. The first relevance calculating means 144 uses the parameters of the number of sessions shown in the above formulas (10) and (11) to formulate a predetermined “item i _a ” and other “ The degree of association f _n (j | a; θ) between items with item i _j ”is calculated.

Here, θ indicates that a parameter is used, and L _n indicates that a function for measuring the parameter distance is used. Equation (16) shows the Kullback Leibler distance for measuring the “distance between Poisson distributions”. It is possible to select items (related items) that are more deeply related to _a predetermined “item i _a ” in ascending order of the value calculated by the equation (16).

<Second relevance calculation means>
The second relevance calculation means 145 reads the time average parameter from the data storage unit 110 and calculates the relevance between any two items among the items to be searched. In the present embodiment, it is assumed that the “time average” indicating the relationship between items follows an exponential distribution. The second relevance calculating means 145 uses the time average parameters shown in the above formulas (12) and (13) to formulate a predetermined “item i _a ” and other “ The degree of association f _t (j | a; θ) between items with the item i _j ”is calculated.

<Third relevance calculation means>
The third relevance calculating means 146 reads out the parameter of the number of items different from the data storage unit 110, and calculates the relevance between any two items among the items to be searched. In the present embodiment, it is assumed that the “average of the number of different items” indicating the relationship between items follows an exponential distribution. The third relevance calculating means 146 uses the parameter of the number of different items shown in the above formulas (14) and (15), and by using the formula (18), a predetermined “item i _a ” The degree of association f _d (j | a; θ) between items with “item i _j ” is calculated.

In the above-described equations (17) and (18), θ indicates that a parameter is used, and L _t (or L _d ) indicates that a function that measures a parameter distance is used. Further, both equations (17) and (18) indicate the Kullback Leibler distance for measuring “distance between exponential distributions”. Moreover, it is possible to select an item (related item) that is more deeply related to _a predetermined “item i _a ” in ascending order of values calculated by both equations.

(Scoring part)
As shown in FIG. 6, the scoring unit (scoring means) 150 includes a user number predicting means 151 and a ranking list creating means 152.

<User number prediction means>
The user number predicting means 151 predicts the number of users accessing the item corresponding to the input. In the present embodiment, the user number prediction unit 151 uses the number of sessions (n _aj ) counted by the session statistic calculation unit 132 and the time average parameter shown in the above equation (13) according to the equation (19). The number of users accessing “item i _j ” after the time interval t after accessing “item i _a ” indicated by the search condition is predicted. In the present embodiment, the user number predicting unit 151 uses the number of sessions (n _aj ) counted by the session statistic calculating unit 132 and the parameter of the number of different items shown in the above equation (15) to calculate the equation ( 20), after accessing “item i _a ” indicated by the search condition, the number of users accessing “item i _j ” is predicted after an interval d of different item numbers. By using this predicted number of users, it is possible to calculate the transition probability from each item in consideration of the time series.

<Ranking list creation means>
The ranking list creation unit 152 extracts the degree of relevance of each item corresponding to the item as the search condition from the data storage unit 110 as a scoring result, and the distance between the Poisson distribution (or exponential distribution) is ascending. A ranking list is created by rearranging. The relevance is calculated in advance by the first relevance calculation unit 144, the second relevance calculation unit 145, and the third relevance calculation unit 146 and stored in the data storage unit 110.

[Operation of related information retrieval device]
The operation of the related information retrieval apparatus 100 shown in FIG. 1 will be described with reference to FIGS. 9 to 11 (see FIG. 6 as appropriate). 9 to 11 are flowcharts showing operations of the history data processing unit, the calculation processing unit, and the scoring unit shown in FIG.

<Processing to be performed in advance before search-part 1>
As illustrated in FIG. 9, based on the operation of the operator, the related information search device 100 receives the target user's history data by the input processing unit 120 and inputs the history data to the history data processing unit 130 (step S1). . Further, based on the operation of the operator, the related information search device 100 accepts a threshold condition such as a time threshold or an item count threshold by the input processing unit 120 and inputs it to the history data processing unit 130 (step S2). Note that the processing order of step S1 and step S2 is arbitrary.

  Subsequently, the history data processing unit 130 generates a session for each user from the received all history data by the session data generation unit 131 (step S3), and the session statistics calculation unit 132 sets the session statistics as the session statistics. Every time, an item, an item pair, and its frequency are counted (step S4). Then, the history data processing unit 130 uses the time statistic calculation unit 133 to calculate the time average as the statistic between items as the time statistic for the item pair (step S5). Then, the history data processing unit 130 calculates the average of the number of different items as the statistical amount of the item number and the statistical amount between the items for the item pair by the item number statistical amount calculating unit 134 (step S6). Then, the history data processing unit 130 outputs history data statistics to the data storage unit 110 (step S7). Note that the processing order of steps S5 and S6 is arbitrary.

<Process to be performed in advance before search-part 2>
As shown in FIG. 10, based on the operation of the operator (or at a predetermined timing), the related information search apparatus 100 uses the calculation processing unit 140 from the data storage unit 110 to collect the statistics of history data. The number of sessions is read out, and a parameter for the number of sessions is calculated (step S11). Further, the related information search device 100 reads the time average as the statistical quantity between items from the data storage unit 110 by the calculation processing unit 140 and calculates the parameter of the time average (step S12). Further, the related information search apparatus 100 reads the average of the number of different items as the statistical quantity between items from the data storage unit 110 by the calculation processing unit 140, and calculates the parameter of the number of different items (step S13). Note that the processing order of steps S11 to S13 is arbitrary.

  Then, the calculation processing unit 140 reads the parameter of the number of sessions from the data storage unit 110 by the first relevance calculating unit 144, and calculates the relevance between any two items among the items to be searched. (16) is calculated respectively (step S14). In addition, the calculation processing unit 140 reads the time average parameter from the data storage unit 110 by the second relevance calculation unit 145, and similarly calculates the relevance between any two items by the above-described equation (17). Calculate (step S15). In addition, the calculation processing unit 140 reads out the parameter of the number of different items from the data storage unit 110 by the third relevance calculating unit 146, and similarly calculates the relevance between any two items according to the equation (18) described above. Each is calculated (step S16). Note that the processing order of steps S14 to S16 is arbitrary.

<Processing at the time of search request>
As illustrated in FIG. 11, based on the operation of the operator, the related information search apparatus 100 receives an item ID as a search condition by the input processing unit 120 and inputs the item ID to the scoring unit 150 (step S <b> 21). When the item ID as the search condition is stored in the data storage unit 110, the scoring unit 150 reads the stored item ID. Then, the scoring unit 150 accesses the item corresponding to the input by the above formula (19) and formula (20) using the statistics and parameters read from the data storage unit 110 by the user number predicting unit 151. The number of users to be predicted is predicted (step S22). Then, the scoring unit 150 uses the ranking list creation unit 152 to extract, from the data storage unit 110, the degree of relevance that matches the item (search condition) corresponding to the input, and rearranges the scoring results. Each ranking list is created (step S23). Then, the scoring unit 150 outputs the ranking list to an output device (not shown) via the output processing unit 160 (step S24).

  Note that the related information search apparatus 100 can also be realized by executing a related information search program that causes a general computer to execute the above-described steps. This program can be distributed via a communication line, or can be written and distributed on a computer-readable recording medium such as a CD-ROM or a flash memory.

  According to the present embodiment, the related information search device 100 uses the number of sessions including item pairs as the time context of each user's action history before executing the search, thereby The degree of association between items defined in consideration of time series changes can be calculated in advance. Thereby, an item can be searched with the relevance to the same preference, and the number of access users for each item can be predicted in time series. Further, since the related information search device 100 uses the relevance between items unlike the conventional method, it is possible to effectively realize personalization by recommending items that match the user's preference as compared with other methods. Furthermore, the related information search device 100 can similarly perform related information search processing without depending on the format of the user history data and the type of action history.

  As mentioned above, although embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning. For example, in the present embodiment, the history data processing unit 130 has been described as the best mode in which “number of sessions”, “time average”, and “average of the number of different items” are all used as statistics based on user history data. However, the present invention is not limited to this, and the effect of the present invention can be obtained as long as at least the “number of sessions” can be used.

  The processing for obtaining the parameter of the index in the calculation processing unit 140 is also the same. Furthermore, the calculation processing unit 140 has been described as the best mode using all parameters of “number of sessions”, “average of time”, and “average of the number of different items” when calculating the relevance, but the present invention is not limited to this. If at least the parameter “number of sessions” can be used, the effect of the present invention is obtained. Moreover, in the scoring part 150, although the user number prediction means 151 shall be predicted using both Formula (19) and Formula (20), either may be sufficient and the user number prediction means 151 is essential. It is not a configuration.

  In addition, all parameters and relevance of “number of sessions”, “time average”, and “average of different items” are obtained and stored in the data storage unit 110, according to the user's request and the user's purpose. Thus, the related item may be searched using at least one of three different viewpoints. Thus, when the statistics and the degree of association of each parameter, session, etc. are obtained and stored in the data storage unit 110, the history processing such as dividing the sequence into sessions is not repeated from the beginning. A plurality of different viewpoints can be searched at a high speed for each search input condition.

Further, in the present embodiment, the degree of association between items is calculated in advance in the calculation processing unit 140 before the search processing. In this case, for example, the above-described equations (11) and (13) are used. Alternatively, a default value can be determined as μ shown in Expression (15), and the degree of association between items for the target history data can be obtained based on the default value. Thereafter, when the user who searches using the related information search apparatus 100 sets the parameter μ to an arbitrary value, the parameters λ _{n, aj} , λ _{t, aj} , λ _{d, aj} are recalculated, and The relevance shown in the equation (16), the equation (17), or the equation (18) can be calculated again. Thereby, a suitable related item according to a user's purpose can be searched.

  Moreover, the related information search device 100 is not limited to outputting a plurality of related items in response to an input of one item ID, and may output only one related item that is most related. Furthermore, one or more related items can be output in response to input of a plurality of item IDs based on conditions such as “AND” and “OR”. In this case, for example, the product of the individual relevance levels can be used for the “AND” condition, and the sum of the individual relevance levels can be used for the “OR” condition.

  In the present embodiment, the related information search device 100 is included in a search service (search engine or EC site) system, but can also be implemented as a system independent of the search service system. In this case, the related item ranking list can be transmitted simultaneously to a plurality of search services. In this case, the input processing unit 120 can accept an output result of another search engine, an output result of an EC site, and an item ID.

  Further, although the ranking list is output to an output device (not shown) via the output processing unit 160, the ranking list can be returned to an independent search service. Further, the present invention is applicable not only to the search service shown in the present embodiment but also to EC site product search and information filtering.

  Moreover, the apparatus which comprises the relevant-information search apparatus 100 is not limited to 1 unit | set, You may distribute and distribute a function to several apparatus. For example, the history data processing unit 130, the calculation processing unit 140, and the scoring unit 150 may be configured as separate devices. As a result, the load on each device is distributed, and high-speed search processing can be realized.

It is a lineblock diagram showing typically the related information search device concerning the embodiment of the present invention. It is explanatory drawing which shows the outline | summary of the process of the history data in the related information search device shown in FIG. It is explanatory drawing which shows the outline | summary of the search process in the related information search device shown in FIG. It is explanatory drawing which shows the outline | summary of the log | history data of a some user. It is explanatory drawing which shows the outline | summary of operation | movement of the relevant-information search apparatus shown in FIG. It is a block diagram which shows an example of a detailed structure of the related information search device shown in FIG. FIG. 7 is an explanatory diagram of processing of the time statistic calculation unit shown in FIG. 6, where (a) shows the case where the reference direction is backward, and (b) shows the case where the reference direction is forward and backward. 7A and 7B are explanatory diagrams of processing of the item number statistic calculation unit illustrated in FIG. 6, in which FIG. 6A illustrates a case where the reference direction is backward and FIG. 6B illustrates a case where the reference direction is front and rear. It is a flowchart which shows operation | movement of the log | history data processing part shown in FIG. It is a flowchart which shows operation | movement of the calculation process part shown in FIG. It is a flowchart which shows operation | movement of the scoring part shown in FIG.

Explanation of symbols

100 Related Information Retrieval Device 110 Data Storage Unit 120 Input Processing Unit 130 History Data Processing Unit (History Data Processing Means)
131 Session data generation means 132 Session statistics calculation means 133 Time statistics calculation means 134 Item number statistics calculation means (object number statistics calculation means)
140 Calculation processing unit (calculation processing means)
141 Session number calculating means 142 Time average calculating means 143 Different item number calculating means (different object number calculating means)
150 Scoring part (scoring means)
144 First relevance calculating means 145 Second relevance calculating means 146 Third relevance calculating means 151 User number predicting means 152 Ranking list creating means 160 Output processing unit

Claims (8)

Searching for at least one of the objects to be analyzed based on the object and the index indicating the relationship between the objects obtained by analyzing the history data indicating the action history of the user performing the action of accessing the object in advance As a condition, a related information search device that searches related objects related to the search condition as related information from among other objects that are the subject of the analysis,
Individual session data for each user represented by a set of user identification information, object identification information, and access time information indicating the time when the user performed an action to access the object, from the history data to be processed From the generated session data, the number of sessions in which a predetermined object appears, the number of sessions in which two predetermined objects whose access order is specified in the object pair appear, and the predetermined object as an object pair History data processing means for obtaining the number of sessions as one of the above as statistics of the history data,
A calculation processing means for calculating a parameter using the number of sessions obtained as a statistic of the history data as the index, and calculating a degree of association between any two objects based on the calculated index; ,
Input processing means for receiving the search condition;
At least a related object that is most relevant to the object indicated by the search condition among the other objects that are subject to the analysis based on the relevance degree calculated in advance with respect to the object indicated by the accepted search condition. A scoring means for generating a ranking list including:
Output processing means for outputting the ranking list;
A related information search device comprising: The history data processing means includes
Time statistic calculation means for obtaining, as a time statistic, an average value of time indicating a difference in access time to any two objects from the number of sessions obtained as the statistic of the history data and the individual session data With
The calculation processing means includes:
Based on the average value of the time obtained as the time statistic, a time average calculation means for calculating a parameter indicating the time average as the index,
The related information search device according to claim 1, further comprising: a second relevance calculating unit that calculates a relevance between any two objects based on the calculated time average parameter. . The history data processing means includes
Indicates the number of different objects accessed from when the user accesses one of the object pairs until the other is accessed, based on the number of sessions determined as the statistics of the history data and the individual session data An object number statistic calculating means for obtaining an average number of objects as an object number statistic;
The calculation processing means includes:
Based on the average value of the number of objects obtained as the object number statistic, a different object number calculating means for calculating a parameter indicating the average of the different number of objects as the index,
The association according to claim 1, further comprising: a third association degree calculating unit that calculates an association degree between any two objects based on the calculated parameter indicating the average number of different objects. Information retrieval device. Searching for at least one of the objects to be analyzed based on the object and the index indicating the relationship between the objects obtained by analyzing the history data indicating the action history of the user performing the action of accessing the object in advance As a condition, a related information search method for a related information search device that searches related objects related to the search condition as related information from among other objects that are the subject of the analysis,
The related information retrieval device includes:
An input processing unit, a history data processing unit, a calculation processing unit, a scoring unit, and an output processing unit;
The history data processing means includes
Receiving the history data of interest;
From the received history data, individual session data for each user represented by a set of user identification information, object identification information, and access time information indicating a time at which the user performs an action to access the object. Generating step;
From the generated session data, the number of sessions in which a predetermined one object has appeared, the number of sessions in which a predetermined two objects in which the access order is specified in the object pair has appeared, and the predetermined object are included as one of the object pairs A step of obtaining the number of sessions as a statistic of the history data,
The calculation processing means calculates a parameter using the number of sessions obtained as a statistic of the history data as the index;
Executing a degree of relevance between any two objects based on the calculated index,
The input processing means executes a step of accepting the search condition,
The scoring means includes
At least a related object that is most relevant to the object indicated by the search condition among the other objects that are subject to the analysis based on the relevance degree calculated in advance with respect to the object indicated by the accepted search condition. Perform the steps to generate a ranking list containing:
The output processing means executes a step of outputting the ranking list. The history data processing means includes a time statistic calculation means,
The time statistic calculation means calculates an average value of time indicating a difference in access time to any two objects from the number of sessions obtained as the statistic of the history data and the individual session data. Perform the step you want as a quantity,
The calculation processing means includes a time average calculation means and a second relevance calculation means,
The time average calculation means executes a step of calculating a parameter indicating a time average as an index at the time of the search based on the average value of the time obtained as the time statistic.
The said 2nd relevance degree calculation means performs the step which calculates the relevance degree between arbitrary two objects based on the parameter which shows the calculated time average, respectively. Related information search method. The history data processing means includes object number statistic calculation means,
The object number statistic calculating means is configured to determine whether the user accesses one of the object pairs after accessing the other from the number of sessions obtained as the statistic of the history data and the individual session data. The step of obtaining the average value of the number of objects indicating the number of different objects accessed as an object number statistic,
The calculation processing means includes different object number calculation means and third relevance calculation means,
The different object number calculation means executes a step of calculating a parameter indicating an average of different objects as an index at the time of the search based on the average value of the object numbers obtained as the object number statistic.
5. The third relevance calculating means executes a step of calculating relevance between any two objects based on the calculated parameter indicating an average of the number of different objects, respectively. The related information search method described in 1.   The related information search program for implement | achieving the function of the related information search apparatus as described in any one of Claim 1 thru | or 3 with a computer.   A computer-readable recording medium on which the related information retrieval program according to claim 7 is recorded.

Images