JP5216895B2 - Log processing apparatus and operation method thereof - Google Patents

Description

  The present invention relates to a log processing apparatus and an operation method thereof.

  Collaborative filtering is a technology that, in a broad sense, accumulates preference information of many users and recommends content to relevant users using information on other users who have similar preferences to a certain user (user).

  This technology can be classified into a type between users and a type between items (also referred to as contents). The inter-user type is a technique that first finds a user (referred to as a similar user) whose preference pattern is similar to the user receiving the recommendation, and sets an item group preferred by the similar user as a recommended candidate. For implementation, the degree of similarity between users is represented by the correlation coefficient (Pearson correlation, rank correlation, etc.) of the evaluation given to the same item, and users with high similarity are extracted for preference prediction. The evaluation value for the item is weighted with the similarity to the user, and the weighted average value of the evaluation values is used as a prediction. Recommendations are made as recommended items from those with large predicted values. Since there is a restriction on the size of the display screen, an item having a small predicted value is deleted, or the upper 3 to 10 items are displayed.

  Also, in the item-to-item type, it is assumed that items that receive similar evaluations by various users are similar, and the assumption that users are interested in similar items of the items they are interested in is the closest to the user. Recommend items that are similar to items in your usage history. As an implementation, the similarity between items is measured based on the co-occurrence of the user of the item, and the similar item of the item in the latest usage history of the user is recommended.

  Details of these techniques are described in Non-Patent Document 1.

Toshihiro Kamijo, “Algorithm of Recommended System (2),” Journal of Japanese Society for Artificial Intelligence Vol.23, No.1, pp.89103, January 2008

  These recommended methods can be used in the case of the same number of users, the number of contents, and the frequency of use even when recommendations are closed within the service or between services (hereinafter referred to as cross-service recommendations). Then, there are the following problems.

  When using the inter-user type recommendation method, the weighted average value of the evaluation values of items is used, so the evaluation value of an item belonging to a service with low usage frequency is used when the usage frequency varies greatly between services. There is a problem that it is not displayed as a recommendation because it is considered to have a large value compared to items belonging to services with a high price.

  The same applies when using the item-to-item recommendation method, and in order to measure the similarity between items based on the co-occurrence of the user who uses the item. There is a problem that items belonging to a low service are unlikely to appear, and a problem that recommendations between different services such as “recommend content β belonging to a service B from a content α belonging to a service A” is unlikely to occur. there were.

  These problems are known to be caused by an imbalance in the number of logs per service.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a log processing apparatus and an operation method thereof for eliminating the imbalance in the number of logs for each service used for content recommendation. Is to provide.

  In order to solve the above-described problem, the first aspect of the present invention recommends content to a destination user when one or more series composed of one or more contents can be transmitted in each of a plurality of services. A log processing device for processing the log accumulated in the transmission, the transmission date and time of the transmission indicating that any content of any series of any service has been transmitted for any user, Log accumulation unit for accumulating logs including the content ID of the content, the service ID of the service, the series ID of the series, and the user ID of the user, and a set of logs having more than a predetermined number of logs For each set of logs, for each set of user ID and series ID, for each set of logs. A log set that is a set of logs including a series ID is searched from the log storage unit, and for each log set, a predetermined number of logs are selected from the log set in order from the newest transmission date and selected. A log processing apparatus including a log deletion unit that deletes a log that has not been performed is used as a solution means.

  For example, the log processing apparatus includes a coefficient calculation unit that obtains a coefficient indicating an imbalance in the number of logs in a set of logs including the service ID for each service ID, and the log deletion unit has the coefficient determined in advance. If it is larger than the threshold, a set of logs having a number of logs larger than a predetermined number is searched from the log storage unit, and for each set of logs, for each set of user ID and series ID, the set of users A log set that is a set of logs including an ID and a series ID is searched from the log storage unit, and for each log set, a predetermined number of logs are selected in order from the newest transmission date and time from the log set. Delete logs that were not selected with.

  For example, the log processing apparatus includes a coefficient calculation unit that obtains a coefficient indicating an imbalance in the number of logs in a set of logs including the service ID for each service ID, and the log deletion unit has the coefficient determined in advance. If it is greater than the threshold value, two log sets having different numbers of logs are searched from the log storage unit, and the period length from the most recent transmission date / time to the oldest transmission date / time of the log set having a small number of logs is determined. And obtaining a period length obtained by multiplying the period length by a predetermined coefficient, detecting the latest transmission date and time from the set of logs having a large number of logs, and detecting the detected transmission date and time for each transmission date and time of the log set. Whether or not it is included in the period up to the time point that has been traced back by the determined period length, and the log including the transmission date and time determined to be included in the period is selected and the log that has not been selected is collected in the log collection To remove from.

  The second aspect of the present invention performs processing of logs accumulated to recommend content to a destination user when one or more series composed of one or more contents can be transmitted in each of a plurality of services. A log processing apparatus operating method, wherein the log processing apparatus transmits a transmission date and time of transmission indicating that any content in any series of any service has been transmitted for any user, A log storage unit that stores a log including the content ID of the content, the service ID of the service, the series ID of the series, and the user ID of the user, and the operation method is determined in advance by the log processing device; The log accumulation unit is searched for a set of logs having a larger number of logs, and a user ID and a log are collected for each set of logs. For each group of IDs, a log set that is a set of logs including the user ID and series ID of the group is searched from the log storage unit, and a predetermined number of log sets are obtained from the log set for each log set. The log processing apparatus operating method is characterized in that logs are selected in order from the most recent transmission date and time, and logs that have not been selected are deleted.

  For example, the log processing device obtains a coefficient indicating an imbalance in the number of logs in a log set including the service ID for each service ID in advance, and the log processing device has the coefficient larger than a predetermined threshold. In this case, a set of logs having a number of logs larger than a predetermined number is searched from the log storage unit, and for each set of logs, for each set of user ID and series ID, the set of user ID and series A log set that is a set of logs including an ID is searched from the log storage unit, and for each log set, a predetermined number of logs are selected from the log set in order from the newest transmission date and time. Delete logs that did not exist.

  For example, the log processing device obtains a coefficient indicating an imbalance in the number of logs in a set of logs including the service ID for each service ID, and the log processing device determines that the coefficient is greater than a predetermined threshold. In the case where the number of logs is large, two log sets having different numbers of logs are searched from the log storage unit, and the period length of the period from the most recent transmission date to the oldest transmission date of the log set having a small number of logs is obtained, The period length obtained by multiplying the period length by a predetermined coefficient is obtained, the latest transmission date / time is detected from the set of logs having a large number of logs, and the respective transmission date / time of the log set is obtained from the detected transmission date / time. Whether or not it is included in the period up to the point of time that is traced back by the length of the selected period, and the log including the transmission date and time determined to be included in the period is selected and the log that has not been selected is the set of logs We want to delete.

  According to the present invention, it is possible to eliminate an imbalance in the number of logs for each service used for content recommendation.

It is a figure which shows the structure of the communication system using the log processing apparatus which concerns on this Embodiment. It is a figure which shows an example of the content of a history information database. It is a figure which shows an example of the content of a meta information database. It is a block diagram which shows schematic structure of the recommendation server. It is a flowchart which shows the flow of the process in the recommendation server. 3 is a diagram illustrating an example of contents of a log storage unit 11. FIG. It is a figure which shows an example of the log deletion by the log deletion part.

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

  FIG. 1 is a diagram showing a configuration of a communication system using the log processing apparatus according to the present embodiment.

  1, the recommendation server 1 is connected to the content servers 21, 22,... And the user terminals 31, 32,. Each user terminal is used by an individual user who uses the user terminal. The log processing apparatus according to the present embodiment is configured in the recommendation server 1.

  In response to a request from a user terminal, the content server 21 performs, for example, a service for transmitting book data to the user terminal.

  In response to a request from a user terminal, the content server 22 performs, for example, a service for transmitting video data to the user terminal.

  Here, the data of one book and the data of one video are both referred to as contents.

  In the present embodiment, for example, a series composed of two or more contents such as the first volume, the second volume,. Also, for example, a configuration such as video data of a movie and data of its sequel is also called a series. Here, the single content is also one series.

  The recommendation server 1 obtains a content that is considered to be preferred by each user based on data acquired from the content server, and notifies the corresponding user terminal of the content, that is, recommends the content.

  Each content server includes a content database 101 that stores content that can be transmitted, a history information database 102 that stores history information indicating the history of content transmission, and a meta information database that stores meta information indicating the relationship between the content and the series. 103.

  FIG. 2 is a diagram illustrating an example of the contents of the history information database 102.

  The history information database 102 stores history information. Each history information includes the transmission date and time when the corresponding content is transmitted, identification information indicating the content (hereinafter referred to as a content ID), and a user who uses the destination user terminal. Identification information (hereinafter referred to as user ID).

  FIG. 3 is a diagram illustrating an example of the contents of the meta information database 103.

  The meta information database 103 stores meta information about each content, and each meta information has a content ID of the corresponding content and identification information (hereinafter referred to as a series ID) indicating a series including the content.

  FIG. 4 is a block diagram illustrating a schematic configuration of the recommendation server 1.

  The recommendation server 1 includes a log storage unit 11 configured based on history information and meta information acquired from each content server, and a Gini coefficient calculation for obtaining a Gini coefficient GC indicating an imbalance in the number of logs in the log storage unit 11. Unit 12, a log deletion unit 13 that deletes a log from log storage unit 11, and a content recommendation unit 14 that notifies a user terminal of content that is considered to be preferred by the user.

  FIG. 5 is a flowchart showing the flow of processing in the recommendation server 1.

  The recommendation server 1 acquires history information and meta information from each content server, and configures the log storage unit 11 (S1).

  FIG. 6 is a diagram illustrating an example of the contents of the log storage unit 11.

  The log accumulating unit 11 accumulates a log corresponding to the acquired history information, and each log includes a transmission date and time included in the corresponding history information, a content ID included in the history information, and an acquisition source content server Identification information (hereinafter referred to as a service ID) indicating a service corresponding to, a series ID included in the meta information including the content ID, and a user ID included in the history information.

  Returning to FIG. 5, the Gini coefficient calculation unit 12 calculates the Gini coefficient GC based on the log storage unit 11 (S3).

Here, the Gini coefficient calculation unit 12 first calculates the average difference MDF according to Equation (1).

  Here, n is the number of services (the number of content servers), xi is the number of logs including the service ID of service i, and xj is the number of logs including the service ID of service j.

Next, the Gini coefficient calculation unit 12 calculates the average value μ using Equation (2).

  Here, n is the number of services (the number of content servers), x1, x2,... Are the number of logs including the service ID of the first service (for example, the service corresponding to the content server 21), The number of logs including the service ID of the second service (for example, the service corresponding to the content server 22), ..., the number of logs including the service ID of the nth service.

  Next, the Gini coefficient calculation unit 12 calculates the Gini coefficient GC using Expression (3).

GC = MDF / (2 × μ) (3)
The Gini coefficient GC indicates an imbalance in the number of logs in a set of logs (hereinafter referred to as a set of logs) including the service ID for each service ID. The Gini coefficient GC is included in the range of 0-1. The closer the Gini coefficient GC is to 1, the greater the degree of imbalance, and the closer the Gini coefficient GC to 0, the smaller the degree of imbalance. When there is no imbalance, that is, when the number of logs is equal to each other, the Gini coefficient GC becomes zero.

  Next, the log deletion unit 13 determines whether or not the Gini coefficient GC is larger than a predetermined value (hereinafter referred to as a threshold GCT) (S5).

  If the Gini coefficient GC is greater than the threshold value GCT (S5: YES), the log deletion unit 13 searches the log storage unit 11 for a set of logs having a larger number of logs than a predetermined number (hereinafter referred to as threshold value LT). (S7) For each set of logs, the processes of steps S9 and S11 are performed for each set of user ID and series ID.

  In step S9, the log deletion unit 13 searches the log storage unit 11 for a set of logs (hereinafter referred to as a log set) including the user ID and series ID of the set (S9).

  In subsequent step S11, the log deleting unit 13 selects a predetermined number of logs from the log set in order from the newest transmission date and time, and deletes unselected logs (S11).

  As shown in FIG. 7, in step S11, the log deletion unit 13 sequentially selects and selects three logs equal to a predetermined number 3 from the log set including five logs in order from the newest transmission date and time. Delete the two logs that were not done.

  Returning to FIG. 5, next, the Gini coefficient calculation unit 12 calculates the Gini coefficient GC similarly to Step S <b> 3 (S <b> 13), and the log deletion unit 13 determines that the Gini coefficient GC is greater than the threshold GCT, similar to Step S <b> 5. It is determined whether it is larger (S15).

  If the Gini coefficient GC is greater than the threshold GCT (S15: YES), the log deletion unit 13 determines from the log storage unit 11 a set of logs having the smallest number of logs equal to or less than the threshold LT (hereinafter referred to as a log set LL). One is searched (S17), the period length of the period from the most recent transmission date and time to the oldest transmission date and time of the log set LL is obtained, and a predetermined coefficient belonging to the range of 0 to 1 (for example, , 0.5) is obtained (S19). If the period length of the period from the most recent transmission date / time of the log set LL to the oldest transmission date / time is 30 days and the coefficient is 0.5, the period length T is 15 days.

  Next, the log deletion unit 13 searches the log storage unit 11 for a set of logs having a number of logs larger than the threshold LT (hereinafter, referred to as a log set LM) (S21). The process of S23 is performed.

  In step S23, the most recent transmission date and time (hereinafter referred to as transmission date and time P) is detected from the log set LM, and each transmission date and time of the log set LM is a period from the transmission date and time P to the time point that is back by the period length T. The log including the transmission date / time determined to be included in the period is selected and the log not selected is deleted from the log set LM (S23).

  Next, the Gini coefficient calculation unit 12 calculates the Gini coefficient GC similarly to Steps S3 and S13 (S25), and the log deletion unit 13 determines that the Gini coefficient GC is larger than the threshold GCT, similar to Steps S5 and S15. It is determined whether or not (S27).

  If the Gini coefficient GC is greater than the threshold value GCT (S27: YES), the log deletion unit 13 obtains a value (hereinafter referred to as a threshold value LT2) obtained by multiplying the average value μ calculated in the process of step 25 by a predetermined multiple. (S29) A set of logs having a number of logs larger than the threshold LT2 is searched from the log storage unit 11 (S31), and if there is a set of logs having a number of logs larger than the threshold LT2 (S32: YES), For the set of logs, the processes of steps S9 and S11 are performed for each set of user ID and series ID. After step S11, the process proceeds to step S13.

  If it is determined in step S5, S15, or S27 that the Gini coefficient GC is equal to or less than the threshold value GCT (NO), or if there is no log set having more logs than the threshold value LT2 (S32: NO), for each user, the content recommendation unit 14 uses the log storage unit 11 to obtain the content ID of the content that the user is likely to like in a range of a predetermined maximum number or less, and obtains the content ID. By transmitting the included information to the user terminal corresponding to the user (S33), the content is recommended to the user, and a series of processing ends.

  The content recommendation unit 14 obtains a series ID of a series composed of content that is considered to be preferred by the user, and transmits information including the series ID to the user terminal (S33). You may recommend the content that you make up.

  It should be noted that the technology of Non-Patent Document 1 can be used to obtain content that is considered to be preferred by users. Here, relevance between contents, series, and services is required, and content is determined based on such relevance.

  If the technology is used for the log storage unit 11 (initial log storage unit 11) before deletion by the log deletion unit 13, the number of logs including the service ID of a certain service is the initial log storage unit 11 When the number of logs including other service IDs is extremely small, the content of the service is not recognized as content that the user would like, and therefore is not recommended.

  However, in the log storage unit 11 after deletion by the log deletion unit 13, the number of logs including the service ID of the service is relatively large. Therefore, the content of the service is the content that the user thinks is preferred. Can be recommended.

  Therefore, according to the present embodiment, in the recommendation server 1, the log storage unit 11 in which logs including the transmission date / time, content ID, service ID, series ID, and user ID are stored, and the service ID for each service ID. If the Gini coefficient GC is greater than a predetermined threshold GCT (S5: YES), a predetermined number (S5: YES) is obtained. A set of logs having a larger number of logs than the threshold value LT) is searched from the log storage unit 11 (S7). For each set of logs, for each set of user ID and series ID, the set of user ID and series ID and The log accumulation unit 11 is searched for a log set that is a set of logs including (S9), and for each log set, a predetermined number ( In the example of FIG. 7, the log of “3”) is selected in order from the newest transmission date and time, and the log that has not been selected is deleted (S11). Can eliminate the imbalance in the number of logs.

  If the Gini coefficient GC is larger than the threshold GCT (S15: YES), the log deletion unit 13 searches the log storage unit 11 for two log sets (log sets LL and LM) having different numbers of logs (S17). S21), the period length of the period from the newest transmission date and time to the oldest transmission date and time of the log set LL with a small number of logs is obtained, and the period length (period length T) obtained by multiplying the period length by a predetermined coefficient (S19), the most recent transmission date and time (transmission date and time P) is detected from the log set LM with a large number of logs, and each transmission date and time of the log set is determined from the detected transmission date and time P. It is determined whether or not it is included in a period up to a point back by the length T, a log including a transmission date and time determined to be included in the period is selected, and a log that has not been selected is deleted from the set of logs ( S23) It is possible to eliminate a number of the imbalance of the log of each service.

  In this embodiment, the Gini coefficient GC is used, but any other coefficient may be used as long as it is a coefficient indicating the imbalance of the number of logs for each service. Further, it may be configured such that log deletion processing is performed on a set of logs having a large number of logs without performing condition determination using a coefficient indicating imbalance.

  Further, S9 and S11 may be performed only for a log set corresponding to a predetermined service type.

  The computer program for causing the computer to function as the log processing apparatus according to the present embodiment can be recorded on a computer-readable recording medium such as a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, or a magnetic tape. It can be widely distributed by being transmitted via a communication network such as the Internet.

DESCRIPTION OF SYMBOLS 1 ... Recommendation server 11 ... Log storage part 12 ... Gini coefficient calculation part 13 ... Log deletion part 14 ... Content recommendation part 21, 22 ... Content server 31, 32 ... User terminal 101 ... Content database 102 ... History information database 103 ... Meta information The database

Claims (7)

A log processing device that processes logs accumulated to recommend content to a destination user when one or more series composed of one or more contents can be transmitted in each of a plurality of services,
The transmission date and time of the transmission indicating that any content of any series of any service has been transmitted for any user, the content ID of the content, the service ID of the service, the series of the series A log storage unit for storing a log including an ID and the user ID of the user;
A set of logs having a number of logs larger than a predetermined number is searched from the log storage unit, and for each set of logs, for each set of user ID and series ID, the user ID and series ID of the set are set. A log set that is a set of logs to be included is searched from the log storage unit, and for each log set, a predetermined number of logs from the log set are selected in order from the newest transmission date and time, and logs that are not selected And a log deleting unit that deletes the log. A coefficient calculation unit for obtaining a coefficient indicating an imbalance in the number of logs in a set of logs including the service ID for each service ID;
If the coefficient is greater than a predetermined threshold, the log deletion unit searches the log storage unit for a set of logs having a larger number of logs than a predetermined number, and for each log set, a user ID and For each set of series IDs, a log set, which is a set of logs including the user ID and series ID of the set, is searched from the log storage unit, and a predetermined number of logs from the log set for each log set The log processing apparatus according to claim 1, wherein the logs are selected in order from the most recent transmission date and time, and logs that have not been selected are deleted. A coefficient calculation unit for obtaining a coefficient indicating an imbalance in the number of logs in a set of logs including the service ID for each service ID;
If the coefficient is greater than a predetermined threshold, the log deletion unit searches the log storage unit for a set of two logs having different numbers of logs, and starts from the most recent transmission date and time of the set of logs with a small number of logs. The period length of the period up to the oldest transmission date and time is obtained, the period length obtained by multiplying the period length by a predetermined coefficient is obtained, the latest transmission date and time is detected from the set of logs having a large number of logs, and the set of logs For each transmission date and time, it is determined whether or not it is included in a period from the detected transmission date and time to the time point that is back by the determined period length, and a log including the transmission date and time determined to be included in the period is selected. The log processing apparatus according to claim 1, wherein a log that has not been selected is deleted from the set of logs. An operation method of a log processing apparatus that processes accumulated logs in order to recommend content to a destination user when one or more series composed of one or more contents can be transmitted in each of a plurality of services. There,
The log processing device includes:
The transmission date and time of the transmission indicating that any content of any series of any service has been transmitted for any user, the content ID of the content, the service ID of the service, the series of the series A log storage unit for storing a log including an ID and the user ID of the user;
The operation method is as follows:
The log processing device searches the log storage unit for a set of logs having a larger number of logs than a predetermined number, and for each set of logs, for each set of user ID and series ID, the set of users A log set that is a set of logs including an ID and a series ID is searched from the log storage unit, and for each log set, a predetermined number of logs are selected in order from the newest transmission date and time from the log set. And an operation method of the log processing apparatus, wherein the log that has not been selected is deleted. The log processing device obtains a coefficient indicating an imbalance of the number of logs in a set of logs including the service ID for each service ID in advance,
In the case where the log processing device has a coefficient larger than a predetermined threshold, the log processing device searches the log storage unit for a set of logs having a larger number of logs than a predetermined number. And for each set of series IDs, a log set that is a set of logs including the user ID and series ID of the set is searched from the log storage unit, and for each log set, a predetermined number of log sets are obtained from the log set. 5. The operation method of the log processing apparatus according to claim 4, wherein logs are selected in order from the most recent transmission date and time, and logs not selected are deleted. The log processing device obtains a coefficient indicating an imbalance of the number of logs in a set of logs including the service ID for each service ID in advance,
When the log processing device searches for a set of two logs having different numbers of logs from the log storage unit when the coefficient is larger than a predetermined threshold, the most recent transmission date / time of the set of logs having a small number of logs To obtain the period length of the period from the oldest transmission date to the oldest transmission date, determine the period length obtained by multiplying the period length by a predetermined coefficient, detect the latest transmission date and time from the set of logs having a large number of logs, For each transmission date and time of the set, it is determined whether or not it is included in a period from the detected transmission date and time to the time point that is back by the determined period length, and a log including the transmission date and time determined to be included in the period is selected The log processing apparatus operating method according to claim 4, further comprising: deleting a log that has not been selected from the set of logs.   A computer program for causing a computer to function as the log processing device according to claim 1.

Images