JP5519824B1 - Interest analysis method, interest analysis apparatus, and interest analysis program - Google Patents

Abstract

An object of the present invention is to reduce the amount of data required when analyzing user interests and to estimate the amount of data required in advance.
An intermediate result obtained when calculating a user interest score is aggregated for each predetermined aggregation period and stored in a box, and a user interest score is obtained using the intermediate result stored in a box belonging to a history window size. Is calculated. As a result, when only the history window size is to be analyzed or when weighting is performed according to time, it is possible to reduce the data amount of intermediate results to be saved and to estimate the data amount in advance.
[Selection] Figure 1

Description

  The present invention relates to a technique for analyzing a user's interest from a content browsing history.

  There is a demand for a technique for recommending appropriate services and contents in accordance with user behavior and situations. Therefore, the inventors have proposed an interest analysis method and an interest analysis device that can estimate the user's interest with high accuracy by utilizing the scarcity of concept appearance (Patent Document 1).

JP 2013-003797 A

  In Patent Literature 1, when analyzing user interests, the history of all past periods is set as an analysis target, and the interest level of the user is updated using the input history information. If you want to analyze only the history of a certain period in the past, or if you want to weight the history accumulated in the past according to the time difference (elapsed time) from the present, it is necessary to save the intermediate results of the past period . However, there is a problem that a large amount of data is stored as an intermediate result. In addition, there is a problem that the data amount of the intermediate result calculated in the past fixed period cannot be estimated in advance.

  The present invention has been made in view of the above, and when analyzing a user's interest, only a history of a certain period in the past is to be analyzed, or according to a time difference (elapsed time) from the present with respect to a history accumulated in the past. It is an object to reduce the amount of data required for weighting and to make a preliminary estimate of the required amount of data.

An interest analysis method according to a first aspect of the present invention is an interest analysis method for obtaining a user interest score indicating a user's interest in a concept assigned to content, the first content list browsing a plurality of contents as a list, Receiving a second content list selected and browsed from the first content list, a total number of contents in the first content list as a first total, and a processing target in the first content list The number of contents in which the concept appears is the first number of appearances, the total number of contents in the second content list is the second total number, and the number of contents in which the concept appears in the second content list is the second number When the number of appearances is used, the second content list is satisfied under the conditions of the first total number, the first appearance number, and the second total number. Calculating a first probability that the number of contents in which the concept appears is equal to or greater than the second occurrence number and a second probability that is equal to or less than the second occurrence number, and the first probability and the second probability Calculating a feature score by an inverse function of a cumulative distribution function of a standard normal distribution based on the probability of, and calculating an intermediate result of a predetermined aggregation period using the feature score and a weight for the feature score; Storing the intermediate result for each aggregation period, weighting for each aggregation period according to a temporal distance from a predetermined time point, and using the intermediate result weighted for each aggregation period Calculating a user interest score .

  In the above-described interest analysis method, the intermediate results before a predetermined fixed period are aggregated and stored as long-term interest intermediate results, and the intermediate results and weights for each aggregation period aggregated in the aggregation period belonging to the predetermined fixed period are weighted. The user interest score is calculated in combination.

In the interest analysis method, and long-term interest storage area for storing the long-term interest intermediate result, the corresponding storage area for each aggregation period for storing the result of aggregating the intermediate results in the aggregation period to the number of settings of the aggregation period In the step of storing the intermediate results, the intermediate results aggregated in the storage area are stored, and the aggregation period corresponding to the storage area belongs to the predetermined fixed period at a predetermined time. If there is no longer, the intermediate result stored in the storage area is reflected in the long-term interest storage area, the storage area is initialized, and the storage area of the aggregation period newly belonging to the predetermined period It is used as.

  In the above-described interest analysis method, the long-term interest storage area and the storage area are managed as one record in a database.

An interest analysis apparatus according to a second aspect of the present invention is an interest analysis apparatus for obtaining a user interest score indicating a user's interest in a concept assigned to content, wherein the first content list browses a plurality of contents as a list, Receiving means for receiving a second content list selected and browsed from the first content list, and processing the total number of contents in the first content list in the first content list The number of contents in which the target concept appears is the first number of appearances, the total number of contents in the second content list is the second total number, and the number of contents in which the concept appears in the second content list is the second number. The number of occurrences of the second content list under the condition of the first total number, the first appearance number, and the second total number Calculating a first probability that the number of contents in which the concept appears is equal to or greater than the second occurrence number and a second probability that is equal to or less than the second occurrence number, and the first probability and the second probability A calculation means for calculating a feature score by an inverse function of a cumulative distribution function of a standard normal distribution based on the probability of calculating an intermediate result of a predetermined aggregation period using the feature score and a weight for the feature score, An intermediate result update unit that stores the intermediate result for each aggregation period, and uses the intermediate result weighted for each aggregation period, weighted for each aggregation period according to a temporal distance from a predetermined time point And an interest score calculating means for calculating the user interest score .

  An interest analysis program according to a third aspect of the present invention is characterized in that a computer is operated as each part of the interest analysis apparatus.

  According to the present invention, when analyzing the user's interests, the amount of data required when analyzing only a history of a certain period in the past or when weighting according to time is reduced, and the necessary amount of data is estimated in advance. Can be possible.

It is a functional block diagram which shows the structure of the interest analysis apparatus in this Embodiment. It is a figure which shows the example of a list browsing content list. It is a figure which shows the example of a detailed browsing content list. It is a figure which shows the example of the concept system table stored in a concept system / user interest score database. It is a figure which shows the graph of a conceptual system. It is a figure which shows the example of a user interest degree intermediate | middle result table. It is a figure which shows the example of a user interest degree table. It is a figure which shows the example of a presentation content list. It is a flowchart which shows the flow of the process in which the interest analysis apparatus in this Embodiment calculates a user interest score. It is a figure which shows the example of an analysis parameter. It is a figure for demonstrating the log | history window size and a box. It is a flowchart which shows the flow of the process which updates the intermediate result stored in the box.

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

  FIG. 1 is a functional block diagram showing the configuration of the interest analysis apparatus in the present embodiment. The interest analysis apparatus 1 inputs a list browsing content list in which a user browses content as a list, and a detailed browsing content list in which the user browses the content body by selecting from the content list, and is given to each content A user interest score indicating the degree of interest of the user with respect to each concept is calculated to analyze the user's interest. Then, a sorted content score list in which the content is sorted in descending order of the content evaluation score calculated based on the user interest score is output.

  The interest analysis apparatus 1 shown in FIG. 1 includes a history information receiving unit 11, a feature score calculating unit 12, a concept system update processing unit 13, a concept system / user interest score database 14, a presentation content list receiving unit 15, a content database 16, and content. An evaluation processing unit 17 and a sorted content score list transmission unit 18 are provided. Each unit included in the interest analysis device 1 may be configured by a computer including an arithmetic processing device, a storage device, and the like, and the processing of each unit may be executed by a program. This program is stored in a storage device included in the interest analysis apparatus 1, and can be recorded on a recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or provided through a network.

  The history information receiving unit 11 receives a list browsing content list and a detailed browsing content list browsed by the user on the client terminal. The list browsing content list and the detailed browsing content list are generated at the client terminal, transmitted to the content server, and transferred from the content server to the interest analysis device 1.

  The list browsing content list is a list of content that the user has browsed in a list of content titles. FIG. 2 shows an example of the list browsing content list. In the list browsing content list shown in the figure, a cluster ID, a content ID for identifying the content, and a browsing time are described. The browsing time is the time when the title of the content is displayed on the client terminal. When all the titles of content are not displayed at the same time on the client terminal, what is displayed by scrolling the title of the content by the user is described in the list browsing content list. The cluster ID is an identifier uniquely assigned to the list browsing content list and the detailed browsing content list. When the user browses the titles of the contents in a list at a different time (time zone) and displays details, a different cluster ID is given. The conditions for newly assigning the cluster ID under conditions other than the time include when there is no operation for a certain period of time while the list browsing content list is displayed, or when the browsing user (user ID) is switched, the list browsing content list On the other hand, when a narrow search is performed from the viewpoint of the content genre or the like, the browsing mode may be switched in other browsing applications.

  The detailed browsing content list is a list of content that is selected from a list of content browsed by the user and browses the contents (details) of the content body. FIG. 3 shows an example of the detailed browsing content list. The list browsing content list shown in the figure describes a cluster ID, a content ID, and a browsing time. The browsing time is the time when the user selects content from a list of content titles and browses the content body.

  The feature score calculation unit 12 calculates the feature score of each concept assigned to the content by the statistical model of concept selection using the list browsing content list and the detailed browsing content list. Details of the feature score calculation will be described later.

  The concept system update processing unit 13 calculates and updates the user interest score for each concept using the feature score calculated by the feature score calculation unit 12, and the concept in the concept system stored in the concept system / user interest score database 14. The user interest score of the subordinate concept is updated using the relationship information (superordinate concept and subordinate concept).

  FIG. 4 shows an example of a concept system table stored in the concept system / user interest score database 14. The concept system table shown in the figure stores a self-concept ID, a parent concept ID list, and a child concept ID list. All the self-concept IDs in the concept system are stored in association with the parent concept ID and the child concept ID, and a concept structure like a graph of the concept system shown in FIG. 5 is defined. Nodes included in the conceptual system graph shown in FIG. 5 represent concepts, and links represent relationships between concepts. The node value of the conceptual system is the user interest score. In the concept system, the nodes located at the higher level represent the abstract concept, and the nodes located at the lower level represent the specific concept. The concept system is designed and defined in advance by service operators.

  In updating the user interest score, in the present embodiment, the conceptual system update processing unit 13 aggregates the intermediate results obtained when updating the user interest score based on a predetermined aggregation period and stores it for a certain past period. In addition, only the history of a certain period in the past can be analyzed, and weighting is possible according to changes in user interest. Specifically, the history window size is defined as the history window size in the past, the history window size is divided into B boxes with an aggregation period suitable for the forgetting granularity required as a service, and an intermediate result is stored for each box. . The history window size is a total value of B aggregation periods. FIG. 6 shows an example of a user interest level intermediate result table for storing intermediate results. The user interest level intermediate result table holds a user ID and a concept ID, and an intermediate result is stored for each concept ID for each user. In the user interest level intermediate result table, a long-term interest box for storing intermediate results obtained by aggregating intermediate results from the start of aggregation until the history window size is stored, and the history window size is divided into B boxes (three in FIG. 6). There are BOX1 to BOX3 for storing intermediate results obtained by aggregating intermediate results for each aggregation period. The long-term interest box and BOX1 to BOX3 are managed in the concept system / user interest score database 14 as one record. The aggregation period is expressed by a relative value from a certain point in time (for example, the format of the past N days to N + M days) or an absolute value (for example, the format of X month Y day to Z month W day).

  The concept system update processing unit 13 obtains an intermediate result using the feature score and reflects it in the user interest level intermediate result table, and then uses the intermediate result stored in each box of the user interest level intermediate result table. An interest score is calculated, and the user interest score is stored in the user interest degree table. When calculating the user interest score, the box is weighted according to a temporal distance from a predetermined time (for example, the current time). Moreover, you may use combining the intermediate result stored in the long-term interest box. FIG. 7 shows an example of the user interest level table. In the user interest level table, a user interest score is stored for each concept ID for each user.

  The conceptual system / user interest score database 14 stores the conceptual system table, the user interest level intermediate result table, and the user interest level table.

  The presented content list receiving unit 15 receives a presented content list that lists contents to be presented to the user. FIG. 8 shows an example of the presented content list. The presented content list has a content ID, a concept ID / relevance list, a content body, and a content registration time. The concept ID / relationship degree list is a list of combinations of a concept ID of a concept that appears in advance for each content and a value (relevance) indicating the degree of relevance between the concept and the content. For example, the relevance is a value from 0 to 1, and the larger the value, the stronger the relevance. For example, for content related to sports, {baseball concept ID = 1, relevance = 0.5}, {soccer concept ID = 2, relevance = 0.8}. Set the relevance pair.

  The content database 16 stores the presented content list received by the presented content list receiving unit 15 in the content table.

  The content evaluation processing unit 17 calculates a user's evaluation score for the content by probability combination using the user interest score of each concept appearing in the content.

  The sorted content score list transmission unit 18 transmits the content score list sorted in the order of the user's evaluation score.

  Next, the operation of the interest analysis device in the present embodiment will be described.

  FIG. 9 is a flowchart showing a flow of processing in which the interest analysis apparatus according to the present embodiment calculates a user interest score.

  The history information receiving unit 11 receives the user ID (or client terminal ID), the list browsing content list, and the detailed browsing content list, and outputs them to the feature score calculation unit 12 (step S101).

  The feature score calculation unit 12 extracts the concept ID that appears in each content in the detailed browsing content list from the content database 16 and creates an appearance concept list (step S102). Specifically, the concept ID linked to each content ID in the detailed browsing content list is searched from the content table of the content database 16 and listed in the appearance concept list. The appearance concept list is a list of all concept IDs assigned to the contents included in the detailed browsing content list.

  Subsequently, the feature score calculation unit 12 extracts a superordinate concept of each concept ID in the appearance concept list and adds it to the appearance concept list (step S103). Specifically, for each concept in the appearance concept list, a superordinate concept is extracted with reference to the concept system table stored in the concept system / user interest score database 14, and the concept ID of the extracted superordinate concept is used as the appearance concept list. Add to

  Subsequently, the feature score calculation unit 12 calculates the number of appearances for each concept in the appearance concept list for each cluster ID, and extracts analysis parameters necessary for calculating the feature score (step S104). FIG. 10 shows examples of analysis parameters. The analysis parameters shown in the figure include the total content S of the list browsing content list, the total content a of the detailed browsing content list, and the list browsing content list totaled for each concept ID of the appearance concept list. The number of contents N to which the concept ID is assigned and the number of contents n to which the concept ID is assigned in the contents of the detailed browsing content list. It should be noted that the content numbers N and n are calculated for all concept IDs in the appearance concept list including the superordinate concept added in step S103.

Then, the feature score calculation unit 12 calculates the feature score Z for each concept ID using the analysis parameters S, a, N, and n (step S105). Specifically, as shown in the following equation, the detailed browsing is performed under the conditions of the total number S _{j of} the list browsing content list and the number N _j of the contents of the list browsing content list to which the concept i to be processed is assigned. When _aj pieces of content are randomly selected from the content list, the cumulative probability H1 that the number of contents in which the concept i appears is n or more and the cumulative probability H2 that the number of contents in which the concept i appears are n or less are Based on the cumulative probabilities H1 and H2, the feature score _Zij is obtained by the inverse function of the standard normal distribution cumulative distribution function.

  Here, i represents a concept ID and j represents a cluster ID. In the present embodiment, the cumulative probabilities H1 and H2 are obtained by hypergeometric distribution, but other distributions such as binomial distribution and normal distribution may be used.

  Then, the feature score calculation unit 12 generates an update target concept list including a set of the concept ID, the feature score, and the weight w for each cluster. The weight w is a value corresponding to a user's characteristic operation, content browsing time, browsing status, and the like.

  Subsequently, the concept system update processing unit 13 updates the node value of each concept ID described in the update target concept list (step S106).

  Further, the concept system update processing unit 13 extracts a subordinate concept of each concept ID described in the update target concept list, and updates the node value of the subordinate concept (step S107). The feature score used to update the user interest score of the subordinate concept is, for example, the value of the adjacent parent concept that has the largest absolute value of the feature score, the value of the closest parent concept, the average value of the parent concept Alternatively, there is a method of using a value obtained by stochastic coupling. Note that the user interest score updated in step S106 is not updated here.

  Next, processing in which the concept system update processing unit 13 updates the user interest score of each concept will be described.

  In the present embodiment, the concept system update processing unit 13 calculates an intermediate result for each concept in the update target concept list, updates the intermediate result held by the update target box with the calculated intermediate result, The user interest score is updated using the retained intermediate result.

  First, a box for storing intermediate results will be described. In the present embodiment, intermediate results are aggregated by dividing the time every predetermined aggregation period based on a predetermined time point (for example, January 1, 1970, 00:00:00). Store in the prepared box. The number of boxes is determined in advance by the history window size and the length of the aggregation period. For a box whose aggregation period does not belong to the history window size at a predetermined time (for example, processing time), the intermediate result stored in the box is used. It is reflected in the long-term interest box, and is used as an aggregation period box newly initialized and belonging to the history window size.

  FIG. 11 shows the history window size and the state of the box. Time is on the horizontal axis. Each box shown in the figure corresponds to the long-term interest boxes BOX1 to BOX3 in the user interest intermediate result table shown in FIG.

  At the time of FIG. 11A, it is assumed that the time at the time of processing for updating the user interest score belongs to the aggregation period of BOX3. At this time, when data is input and processing for updating the user interest score is performed, the intermediate result of BOX 3 is updated with the calculated intermediate result.

  After that, as shown in FIG. 11B, when the aggregation period of BOX1 does not belong to the history window size at the time of processing, the intermediate result stored in BOX1 is reflected in the long-term interest box, and FIG. BOX1 is used as the next new box after BOX3.

  Next, a process for updating the intermediate result stored in the box will be described. FIG. 12 is a flowchart showing a flow of processing for updating the intermediate result stored in the box.

First, the concept system update processing unit 13 calculates intermediate results tmpX and tmpY using the following formula for each concept ID in the update target concept list (step S200).

  Here, i is a concept ID, n is the number of update processes, Z is a feature score, and w is a weight. The number n of update processes is a value indicating how many times the concept system update process is performed. A series of processes for obtaining the user interest score is performed in cluster ID units. A series of processes performed for each cluster ID is counted as one time. In Patent Document 1, the user interest score is calculated using the intermediate results tmpX and tmpY. In the present embodiment, the intermediate results tmpX and tmpY are aggregated for each box by the following processing, and then the user interest score is calculated.

  Subsequently, the update target box number is determined from the current time (step S201). For example, the box number is 3 in the example shown in FIG. 11A, and the box number is 1 in the examples shown in FIGS. 11B and 11C.

  Then, the variable i indicating the box number is initialized (step S202). In the example shown in FIG. 11, since the box numbers are 1 to 3, the variable i is initialized to 1.

  It is determined whether BOXi indicated by the variable i is within the history window size (step S203). For example, when variable i = 1, BOX1 is within the history window size in the example shown in FIGS. 11A and 11C, and BOX1 is outside the history window size in the example shown in FIG. 11B. . When time elapses and the current time moves to a new box aggregation period, there will be a box outside the history window size. Whether BOXi is outside the history window size can be checked by determining whether the last update date of BOXi is outside the history window size, that is, before the history window size.

  If BOXi is within the history window size (Yes in step S203), it is determined whether BOXi is an update target (step S204). For example, when variable i = 1, BOX1 is an update target in the example shown in FIG.

  If BOXi is not an update target (No in step S204), the variable i is incremented and the process proceeds to the next box (steps S212 and S213). That is, if BOXi is within the history window size and is not an update target, no processing is performed on BOXi.

If BOXi is an update target (Yes in step S204), the intermediate result held by BOXi is updated with tmpX and tmpY (step S205). Specifically, the intermediate results X _BOXi and Y _BOXi held by BOXi are updated by the following equation.

  Then, the last update date of BOXi is updated to the current time (step S211), the variable i is incremented, and the process proceeds to the next box (steps S212 and S213).

  On the other hand, even when BOXi is outside the history window size (No in step S203), it is determined whether BOXi is an update target (step S206).

  If BOXi is not an update target (No in step S206), the intermediate result held by BOXi is reflected in the long-term interest box (step S207), and the intermediate result held by BOXi is reset to 0 (step S208).

  When BOXi is an update target (Yes in step S206), the intermediate result held by BOXi is reflected in the long-term interest box (step S209), and the intermediate result held by BOXi is replaced with tmpX, tmpY (step S210).

When BOXi is outside the history window size, that is, when BOXi no longer belongs to the history window size, as shown in FIG. 11B, the intermediate result held by BOXi is reflected in the long-term interest box and a new box is created. Used as If the intermediate results held by the long-term interest box are LongX and LongY, the long-term interest box is updated by the following equation.

  Note that the process of resetting the intermediate result in step S208 to 0 occurs when a time equal to or longer than the aggregation period has elapsed from the previous processing time and a plurality of boxes are out of the history window size at a time. Since no processing is performed during the aggregation period of the box, this means that the box has no intermediate result.

  The process of replacing the intermediate result of BOXi in step S210 with tmpX, tmpY means that BOXi is used as a box for storing the intermediate result of the aggregation period that newly belongs to the history window size.

  When an intermediate result is set in BOXi in steps S208 and S210, the last update date of BOXi is updated to the current time (step S211), and variable i is incremented (step S212). If all the boxes have not been processed, the process returns to step S203 (No in step S213), and the next box is processed.

  If all the boxes have been processed (Yes in step S213), the process of updating the intermediate result is finished, and the process moves to a process of updating the user interest score.

  Then, the process which updates a user interest score is demonstrated.

When the conceptual system update processing unit 13 updates the intermediate result held by the box, the weights w _t1 and w _t2 set according to the temporal distance from the predetermined time point to the intermediate result held by the box according to the following equation: ,..., W _tn and long-term interest weight w _l are added to update the user interest score Z.

Here, intermediate results Xa, Xb,..., Xn, Ya, Yb,. For example, in the example of FIG. 11 (a), Xa is _{X BOX3,} Ya is _{Y BOX3,} Xb is _{X BOX2,} Yb is _{Y BOX2,} Xc is _{X BOX1,} Yc becomes _{Y BOX1.} When the long-term interest box is not used, shortZ is set as the user interest score Z.

  Next, content evaluation processing will be described.

  The content evaluation processing unit 17 calculates a content evaluation score in response to the request, and generates a sorted content score list in which the content is arranged in descending order of the evaluation score.

  When a concept ID to be analyzed is designated, only a subordinate concept of the concept ID in the concept system is evaluated, and a filtered content list linked with the corresponding concept ID is generated.

Then, the evaluation score EntityZ _x of the content x included in the filtered content list is calculated using the following equation.

Here, Z _i represents the user interest score of concept i, w _i represents the degree of association between content x and concept i, and p represents a set of concept IDs possessed by content x.

  Then, the sorted content score list transmission unit 18 transmits the sorted content score list.

  As described above, according to the present embodiment, the intermediate results obtained when calculating the user interest score are aggregated for each predetermined aggregation period and stored in the box, and stored in the box belonging to the history window size. By calculating the user interest score using the obtained intermediate results, when only the history window size is to be analyzed or when weighting is performed according to time, the data amount of the intermediate results to be saved is reduced, and the data Pre-estimation of quantity can be possible.

  According to the present embodiment, by managing the long-term interest box and BOX1 to BOX3 as one record in the concept system / user interest score database 14, it is possible to reduce the number of reading and updating of the database.

DESCRIPTION OF SYMBOLS 1 ... Interest analysis apparatus 11 ... History information receiving part 12 ... Feature score calculation part 13 ... Concept system update processing part 14 ... Concept system / user interest score database 15 ... Presented content list receiving part 16 ... Content database 17 ... Content evaluation processing part 18 ... Sorted content score list transmitter

Claims (6)

An interest analysis method executed by a computer for obtaining a user interest score indicating a user's interest in a concept given to content,
Receiving a first content list browsing a plurality of contents as a list and a second content list selected and browsed from the first content list;
The total number of contents in the first content list is the first total number, the number of contents in which the concept to be processed appears in the first content list is the first number of appearances, and the total number of contents in the second content list Is the second total number, and the number of contents in which the concept appears in the second content list is the second appearance number, the first total number, the first appearance number, and the second total number Under the condition, a first probability that the number of contents in which the concept appears in the second content list is greater than or equal to the second occurrence number and a second probability that is less than or equal to the second occurrence number are calculated. Calculating a feature score by an inverse function of a cumulative distribution function of a standard normal distribution based on the first probability and the second probability;
Calculating an intermediate result of a predetermined aggregation period using the feature score and a weight for the feature score;
Storing the intermediate result for each aggregation period;
Weighting each aggregation period according to a temporal distance from a predetermined time point, and calculating the user interest score using the intermediate result weighted for each aggregation period;
An interest analysis method characterized by comprising: The intermediate results before a predetermined fixed period are aggregated and stored as a long-term interest intermediate result, and the user interest is combined by weighting and combining the intermediate results for each aggregation period aggregated in the aggregation period belonging to the predetermined fixed period. interest analysis method according to claim 1, wherein the calculating the score. And long-term interest storage area for storing the long-term interest intermediate results, leave number fraction prepared corresponding storage area for each aggregation period for storing the result of aggregating the intermediate results in the aggregation period to the number of settings of the aggregation period ,
The step of storing the intermediate result stores the intermediate result aggregated in the storage area, and if the aggregation period corresponding to the storage area does not belong to the predetermined fixed period at a predetermined time, the storage result The intermediate result stored in the area is reflected in the long-term interest storage area, and the storage area is initialized and used as a storage area for the aggregation period newly belonging to the predetermined period. The interest analysis method according to claim 2 . The interest analysis method according to claim 3, wherein the long-term interest storage area and the storage area are managed as one record in a database. An interest analysis device for obtaining a user interest score indicating a user's interest in a concept assigned to content,
Receiving means for receiving a first content list viewed as a list of a plurality of contents and a second content list selected and browsed from the first content list;
The total number of contents in the first content list is the first total number, the number of contents in which the concept to be processed appears in the first content list is the first number of appearances, and the total number of contents in the second content list Is the second total number, and the number of contents in which the concept appears in the second content list is the second appearance number, the first total number, the first appearance number, and the second total number Under the condition, a first probability that the number of contents in which the concept appears in the second content list is greater than or equal to the second occurrence number and a second probability that is less than or equal to the second occurrence number are calculated. Calculating means for calculating a feature score by an inverse function of a cumulative distribution function of a standard normal distribution based on the first probability and the second probability;
Intermediate result update means for calculating an intermediate result of a predetermined aggregation period using the feature score and a weight for the feature score, and storing the intermediate result for each aggregation period;
An interest score calculation unit that weights each aggregation period according to a temporal distance from a predetermined time point, and calculates the user interest score using the intermediate result weighted for each aggregation period;
An interest analysis apparatus characterized by comprising: 6. An interest analysis program for operating a computer as each part of the interest analysis apparatus according to claim 5 .

Images