JP2010262360A - Apparatus and method for distribution of recommended information - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus and a method for distribution of recommended information which, in order to distribute to a user recommended information such as a content latently desired by the user without omission, accurately generate a separation plane used for evaluating such distribution of recommended information. <P>SOLUTION: Whenever content is transmitted, browsing history information such as the degree of browsing and the degree of non-browsing of each content is updated in a browsing history table 104b. Then, when a recommended request is received by a recommended request reception part 105, a degree of browsing calculation part 106 calculates the cumulative value of the degree of browsing and the degree of non-browsing stored in the browsing history table 104b for each content, and a determination part 107 determines the degree of concern of each content based on the cumulative value. A reliability calculation part 108 calculates the reliability of the determination result based on the absolute value of the degree of concern, and a separation plane generation part 109 generates a separation plane to separate a plurality of content into a satisfaction class and a dissatisfaction class based on the calculated reliability. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a recommendation information distribution apparatus and a recommendation information distribution method.

  In order to distribute recommendation information that matches the user's characteristics, a content-based filtering method is considered as an information filtering method for extracting the recommendation information. In this method, when a user takes an action such as browsing a Web page, the browsed target document is analyzed, and words (keywords) included in the document are extracted. These words are distinguished for each user and recorded as interest / preference information called a user profile. One concept of interest is expressed in the user profile, and it is considered to provide information to the user using this concept. For example, a system is generally known that makes it possible to recommend an appropriate advertisement by matching a keyword assigned to an advertisement product or the like with a user profile (expressed by a keyword).

  Moreover, the technique described in patent document 1 is mentioned as a technique relevant to this. This Patent Document 1 describes a technique for determining and distributing a distribution destination to which recommendation information is distributed and recommendation information to be distributed in accordance with a predetermined recommendation extraction rule.

JP 2007-148878 A

  However, in the background art described above, recommendation information is extracted by performing a matching process, or in the technique described in Patent Document 1, recommendation information is determined based on a predetermined recommendation extraction rule. As a result, there is a risk of leakage of recommendation information.

  That is, in the technique of Patent Document 1, if hobbies / preferences are registered in the recommendation extraction rule, recommendation information suitable for the hobbies / preferences is determined and distributed. Based on this, when purchasing a product, information related to the purchased product will be distributed, but this distributed information is based on direct information such as hobbies / preferences, purchase history, or access history. Therefore, there is a case where content that the user wants is leaked.

  Especially when recommendation information is to be distributed based on the access history, the leakage is considered to be particularly large. That is, when a user browses a site or the like, the browsed site itself is not represented by a single keyword, but is usually represented by a plurality of keywords. For example, when a user searches and browses information related to the movie a, the browsing site introduces the movie a, introduces the original, or introduces the entertainment movie theater. Or introducing performers, and various information (vector information). In this case, when the user is searching for an original book, it is desirable for the user to perform a filtering process specialized for the original book and distribute recommendation information only for that purpose.

  However, with the technology described in Patent Document 1, the entire site being browsed is viewed and information close to the entire site is distributed as recommendation information. Could not deliver.

  Therefore, as a technology for distributing information that is potentially desired in this way, it is determined whether each content is satisfied or dissatisfied based on the presence or absence of browsing by the user, and potentially based on the separation plane generated based on this It is considered to determine satisfaction or dissatisfaction with an unknown content desired.

  However, if satisfaction / dissatisfaction for each content is simply determined based on whether or not the user has browsed, it cannot be accurately determined whether the content is really unsatisfactory for the user. That is, if the user does not browse the content that has not been viewed, the user may be determined to be dissatisfied without fully understanding or misunderstanding the content. For example, the user decides which to browse based on the content summary displayed in the content list, but does not fully understand because it judges the necessity of viewing based on the summary. there is a possibility.

  Therefore, the present invention provides a recommendation information distribution device and a recommendation that can more accurately perform a separation plane for performing evaluation so that recommendation information such as content that the user potentially desires can be distributed without omission. An object is to provide an information distribution method.

  In order to solve the above-described problem, the recommended information distribution apparatus of the present invention transmits a content list indicating a plurality of contents to a client terminal, and when one content is selected from the content list at the client terminal, A content transmission unit that transmits one content, a browsing degree for the one content transmitted by the content transmission unit, and a non-browsing degree for other content other than the one content displayed in the content list; Each time content is transmitted by the content transmission means, browsing history storage means for storing the browsing degree and non-browsing degree, and the browsing degree and non-browsing degree of each content stored in the browsing history storage means , Judge for judging the interest level of each content Reliability calculation means for calculating the reliability of the determination result of each content determined by the determination means based on the browsing degree and non-browsing degree of each content stored in the browsing history storage means, and the reliability Based on the reliability of each content calculated by the degree calculation means, a separation plane generating means for generating a separation plane for separating each content into a satisfaction class and a dissatisfaction class, and a separation plane generated by the separation plane generation means And a recommendation transmission means for transmitting the content evaluated by using as the recommendation information.

  According to the present invention, a content list indicating a plurality of contents is transmitted to a client terminal, and when one content is selected from the content list at the client terminal, the one content is transmitted, and the transmitted one content The browsing degree and the non-browsing degree of other contents other than one content displayed in the content list are stored, and the browsing degree and the non-browsing degree are stored each time the content is transmitted. Based on the browsing degree and non-browsing degree of each content stored here, the interest level of each content is determined, the reliability of the determined determination result of each content is calculated, and the calculated reliability of each content Based on the above, a separation plane for separating each content into a satisfaction class and a dissatisfaction class is generated. And the recommendation information produced | generated using this separation plane is transmitted.

  As a result, unknown content can be separated into a satisfaction class and a dissatisfaction class using a separation plane, and content satisfied by the user can be provided as recommendation information. In particular, by using the reliability of the satisfaction determination result, an accurate separation plane can be generated, and recommendation information with a high degree of interest for the user can be provided.

  The recommended information distribution apparatus according to the present invention further includes a browsing degree calculation unit that increases a non-browsing degree for an unknown content similar to a content that is not selected while being selectable in the content list. It is preferable to determine the interest level of each content using the non-browsing degree increased by the browsing degree calculating means.

  According to the present invention, the degree of non-browsing for unknown content similar to the content that has been selected in the content list but is not selected is increased, and the interest of each content is increased using the increased degree of non-browsing. By determining the degree, the non-browsing degree can be propagated to similar contents, and a more accurate reliability can be calculated with a small number of samples.

  Further, in the recommendation information distribution apparatus according to the present invention, the separation plane generation means may separate the separation plane so as not to allow an error with respect to each content according to the degree of reliability of each content calculated by the reliability calculation means. Is preferably generated.

  According to the present invention, the separation plane is generated so as not to allow an error for each content according to the calculated degree of reliability of each content, so that the reliability is lower than that of the content with high reliability. An error with respect to the content is allowed, and a more accurate separation plane according to the reliability can be generated. That is, even if it is determined that content with low reliability is satisfactory, for example, it is not clear whether it is really satisfactory or not. It is possible to generate a separation plane with high accuracy as a result of not giving importance.

  In the recommendation information distribution apparatus according to the present invention, it is preferable that the separation plane generation unit generates a separation plane using content whose reliability calculated by the reliability calculation unit is a predetermined rank.

  According to the present invention, the separation plane is generated using the content having the calculated reliability up to a predetermined order, and the separation plane is generated ignoring the content having a relatively low reliability. That is, even if it is determined that content with low reliability is satisfactory, for example, it is not clear whether it is really satisfactory or not. It is possible to generate a separation plane with high accuracy as a result of not giving importance.

  According to the present invention, unknown content can be separated into a satisfaction class and an unsatisfied class using a separation plane, and content satisfied by the user can be provided as recommendation information. In particular, by using the reliability of the satisfaction determination result, an accurate separation plane can be generated, and recommendation information with a high degree of interest for the user can be provided.

It is a block diagram which shows the function of the recommendation information delivery apparatus 100 of this embodiment. It is explanatory drawing of the information which the content characteristic vector management table 104a memorize | stores. It is explanatory drawing which shows the specific example of the browsing history table 104b. It is a schematic diagram which shows the hierarchical structure of a content. It is explanatory drawing which shows the score table classified by page ID. It is explanatory drawing which shows the specific example of a satisfaction determination table. It is a schematic diagram when producing | generating a separation plane using the conventional SVM method. It is a schematic diagram when producing | generating a separation plane using a SVM method using determination reliability. It is a flowchart which shows the process of the recommendation information delivery apparatus 100 when producing | generating a separation plane. It is a flowchart which shows a process when transmitting the content isolate | separated into the satisfaction class using the separation plane as recommendation information. It is a conceptual diagram which shows the concept that a non-browsing degree propagates with respect to another similar content.

  Embodiments of the present invention will be described with reference to the accompanying drawings. Where possible, the same parts are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a block diagram illustrating functions of the recommendation information distribution apparatus 100 according to the present embodiment. As shown in FIG. 1, the recommended information distribution apparatus 100 includes a content request receiving unit 101, a content transmitting unit 102 (content transmitting unit, recommended transmitting unit), a content list storage unit 103a, a content storage unit 103b, and a page ID score table 103c. , Content characteristic vector management table 104a, browsing history table 104b, satisfaction determination table 104c, non-browsing history table 104d, recommendation request receiving unit 105, browsing level calculation unit 106 (browsing level calculation unit), determination unit 107 (determination unit), A reliability calculation unit 108 (reliability calculation unit) and a separation plane generation unit 109 (separation plane generation unit) are included.

  The recommended information distribution apparatus 100 includes a CPU, a ROM, a RAM, a communication module, and the like, and the following functions can be realized by operating the CPU according to a program stored in the ROM. .

  The content request receiving unit 101 is a part that receives a content distribution request from a mobile device or the like that is a client terminal.

  The content transmitting unit 102 is a part that transmits the content or content list requested from the client terminal based on the content distribution request received by the content request receiving unit 101. Note that the content is information such as a Web page, and the content list is information that summarizes the outline, title, and the like of the content in a list format. This content list is not stored in advance as a list, but is a piece of information in which summary information for posting in the list is stored for each content and is dynamically generated (listed) as necessary. .

  In addition, when the recommendation request is received by the recommendation request receiving unit 105, the content transmitting unit 102 receives at least content and a content list that are evaluated as satisfactory based on a separation plane generated by the separation plane generation unit 109 described later. One of them is transmitted as recommendation information.

  The content list storage unit 103a is a part that temporarily stores a content list describing an outline of content, a title, and the like when the content is generated.

  The content storage unit 103b is a part that stores content to be distributed.

  The page ID score table 103c is a part that stores a score for browsing degree in association with each page ID. As will be described later, the browsing degree calculation unit 106 can describe the browsing degree in the satisfaction determination table 104c using the page ID score table 103c.

  The content characteristic vector management table 104a is a part that stores content characteristic vectors.

  The content characteristic vector management table 104a is a part that stores content characteristic vectors. For example, the information shown in FIG. 2 is stored. FIG. 2 is an explanatory diagram of information stored in the content characteristic vector management table 104a, in which a parameter of the content characteristic vector is stored for each content ID that specifies the content. For example, when the content ID is C1, the characteristic vector (0, 0,..., 1, 0, 0) is defined for the parameters of finance, travel,. Yes.

  The browsing history table 104b and the non-browsing history table 104d are portions that store browsing history information of browsing content selected and browsed from the content list transmitted by the content transmission unit 102 and non-browsing content that has not been browsed. . FIG. 3 is an explanatory diagram showing specific examples of the browsing history table 104b and the non-viewing history table 104d. As shown in FIG. 3A, the browsing history table 104b stores the browsing date, the content ID indicating the browsed content, and the page ID assigned to the content, as shown in FIG. 3B. The non-browsing history table 104d stores the date and time when the content was not browsed despite being in a browsable state and the content ID indicating the content. Each time content is transmitted by the content transmitting unit 102, the above-described various information is stored in the browsing history table 104b, and the content is not transmitted even though the content is displayed in a browsable state. For the content, the content ID and the date and time of non-browsing (the date and time when the content was not browsed even if it was viewable) are stored in the non-browsing history table 104d.

  Note that the values of the content ID and the page ID in the browsing history information may be determined based on URL information or the like recorded as an access log in the recommendation information distribution apparatus 100. That is, in the recommendation information distribution apparatus 100 or another server apparatus, the content ID and the page ID are stored in association with the URL information, and the content ID and the page ID are based on the URL information described in the access log. May be extracted. Further, the content ID and the page ID may be stored as an access log as they are, and the stored information may be used.

  In addition to storing the access log in the recommendation information distribution apparatus 100, the access log may be stored in a Web server that distributes content, and the access log may be extracted from the Web server as necessary. It is also possible to store the access log and extract the access log.

  Here, the page ID included in each content will be described. The page ID in the present embodiment is information indicating the number of points assigned to each content in order to calculate the browsing level. In order to increase the degree of interest in the content, the page ID should be set so that more points are given to the content located in the lower layer among the content structured in a hierarchical structure. Although it is preferable, of course, it is not limited to this.

  FIG. 4 is a schematic diagram showing the hierarchical structure of content. As shown in FIG. 4, when the content (content ID: 1) is selected by the user from the content list 401, the content request receiving unit 101 reads the content 401a from the content storage unit 103b and sends it to the client terminal. Sent. In this content 401a, page ID: 1 is set. In the page ID score table 103c, as shown in FIG. 5, the page ID: 1 is associated with the score: 1. The page ID score table 103c is stored in the storage unit 103, but is not necessary when the browsing degree is not calculated by the page ID.

  When “view map” described in the content 401a is selected by the user, the content 402a, which is a content map screen, is extracted and transmitted. When the user selects “Register to Favorites” described in the content 401a, the content 403a, which is a favorite registration screen, is extracted and transmitted.

  Then, the content transmitting unit 102 transmits the selected content, and the page ID corresponding to each selected content is described in the browsing history table 104b in association with the content ID. For other contents that are displayed in the content list so as to be selectable but are not selected, the content ID of the other contents is described in the non-viewing history table 104d in order to calculate the non-viewing degree.

  The satisfaction determination table 104c is a part that stores information for satisfaction determination generated by the browsing degree calculation unit 106, the determination unit 107, and the reliability calculation unit 108 separately for each user (client terminal). Further, the browsing degree, the non-browsing degree, the interest degree, the satisfaction determination result, and the determination reliability are stored in association with the content ID indicating the content. A specific example is shown in FIG. FIG. 6 is an explanatory diagram showing a specific example of the satisfaction determination table 104c. As shown in FIG. 6, the content ID, the browsing level, the non-browsing level, the interest level, the satisfaction determination result, and the determination reliability are stored in association with each other. The satisfaction determination table 104c is prepared for each user (client terminal).

  As will be described later, the browsing degree and the non-browsing degree are calculated by the browsing degree calculation unit 106 and described in the corresponding part of the satisfaction determination table 104c. The interest level is calculated by subtracting the non-browsing level from the browsing level by the determination unit 107, and the calculation result is described. Similarly, the satisfaction determination result describes the result determined according to the degree of interest calculated by the determination unit 107. The determination reliability is calculated by obtaining an absolute value of the interest level by the reliability calculation unit 108.

  The recommendation request receiving unit 105 is a part that receives a recommendation request that is a distribution request for recommendation information from a client terminal. When the recommendation request receiving unit 105 receives the recommendation request from the client terminal, the content evaluation unit 110 satisfies a plurality of contents according to the degree of interest and the reliability calculated based on the degree of browsing and non-viewing of the content. An instruction is output so that the content is separated according to a separation plane for separating into a class or a dissatisfaction class, and the content located in the satisfaction class is used as recommendation information.

  The content evaluation unit 110 is a part that evaluates content as recommendation information. Specifically, the content evaluation unit 110 includes a browsing degree calculation unit 106, a determination unit 107, a reliability calculation unit 108, and a separation plane generation unit 109. Yes.

  When the recommendation request receiving unit 105 receives a recommendation request from the client terminal, the browsing degree calculation unit 106 is based on information stored in the browsing history table 104b and the non-browsing history table 104d, and the browsing degree of the user of the client terminal It is the part that calculates the non-browsing degree.

  For example, when calculating the browsing level for each content, the browsing level calculation unit 106 converts the page ID stored in the browsing history table 104b into a score using the page ID score table, and converts the score into the content. The total value is calculated for each ID and the accumulated value is calculated. On the other hand, the browsing degree calculation unit 106 counts the number of the same content IDs stored in the non-browsing history table 104d (the number of non-browsing times). The browsing degree calculation unit 106 calculates a numerical value obtained by multiplying a predetermined coefficient by the number of non-browsing times as a non-browsing degree in order to accumulate a predetermined score per non-browsing number.

  The browsing degree and the non-browsing degree calculated in this way are described in the browsing degree column and the non-browsing degree column of the satisfaction determination table 104c as described above.

  In addition, when calculating the non-browsing degree, the browsing degree calculating unit 106 adds a non-browsing count to content similar to the content that has not been browsed to increase the non-browsing degree. Good. For example, when calculating the degree of non-browsing of a certain content, if the content similar to that content is in a state where it can be browsed but has not been browsed, the number of times is calculated based on the non-browsing history table 104d. . Then, the numerical value obtained by multiplying the number of times by a predetermined weighting coefficient is set as the degree of non-viewing propagated from the similar content. When there are a plurality of similar contents, the non-browsing degrees are added up by that amount, and the total non-browsing degree considering the propagated non-browsing degree is stored in the satisfaction determination table 104c.

  Therefore, based on the content characteristic vector stored in the content characteristic vector management table 104a, the similarity is obtained by calculating the inner product, and other contents having a predetermined similarity (that is, similar contents) Since there is a high possibility of not being browsed in the same manner as the content that has not been browsed, the evaluation that the page has not been browsed is propagated despite being in a viewable state. This makes it possible to accurately determine the satisfaction class or the dissatisfaction class with a small number of samples.

  The concept will be described with reference to FIG. FIG. 11 is a conceptual diagram showing the concept of propagation of the non-browsing degree in the N-dimensional feature space. When the content C is not browsed as shown in FIG. 11, the non-viewing degree is added by 2 for the content C4 and C5 similar to the content C. On the other hand, the non-browsing degree is not added for the dissimilar contents C1 to C3. As a result, as described above, the satisfaction class and the dissatisfaction class can be determined with a small number of samples.

  The determination unit 107 calculates an interest level based on the browsing level and the non-browsing level calculated by the browsing level calculation unit 106, and determines whether the corresponding content is a satisfaction class or a dissatisfaction class based on the interest level. It is a part to do. For example, the determination unit 107 calculates the interest level by subtracting the non-viewing level from the browsing level, and if the level of interest takes a positive value, the determination result that the content is a satisfaction class is obtained. The result is described in the satisfaction determination table 104c.

  The reliability calculation unit 108 is a part that calculates the determination reliability based on the degree of interest. Specifically, the reliability calculation unit 108 calculates the absolute value of the interest level and calculates the value as the determination reliability.

  Based on the determination reliability calculated by the reliability calculation unit 108, the separation plane generation unit 109 generates a separation plane for separating a plurality of contents stored in the content storage unit 103b into a satisfaction class and a dissatisfaction class. This is the part to be generated. In the present embodiment, the separation plane generation unit 109 generates a separation plane using an SVM (Suport Vector Machine) method.

Here, a specific method for generating a separation plane will be described. The conventional SVM algorithm calculates a vector w representing a separation plane that minimizes L (w, ξ) using the following equation (1).

Here, 1 / ‖w‖ is a distance between the identification plane and the separation plane, ξ _i is a parameter indicating a distance indicating how far the sample data i has entered beyond the identification plane, and γ is a constant. N is the total number of contents. Thus, a vector w representing the separation plane that minimizes the total error Σξ _i is calculated while maximizing the distance (= margin) 1 / ‖w‖ between the identification plane and the separation plane. The identification plane is a plane that represents the inner and outer boundaries of the satisfaction class (dissatisfaction class), and is a plane parallel to the separation plane.

  FIG. 7 is a schematic diagram of the separation plane w into which the satisfaction class and the dissatisfaction class are separated using the conventional SVM method described above. The separation plane obtained here allows errors for some contents, thereby minimizing the error while maximizing the distance between the identification plane and the separation plane. For example, the content A should originally be positioned in the unsatisfactory class, but protrudes outside the unsatisfactory class, and an error is allowed. In addition, the content B should originally be positioned in the satisfaction class, but protrudes outside the satisfaction class, and an error is allowed.

  On the other hand, in the extended SVM method in the present embodiment, the allowable range of this error is changed based on the determination reliability for determining the interest level determined for each content.

Specifically, in the extended SVM method, a vector w representing a separation plane that minimizes L (w, ξ) is calculated using the following equation (2).

Again, as described above, 1 / ‖w‖ is the distance between the identification plane and the separation plane, ξ _i is the distance indicating how far the sample data content i has entered the opposite side beyond the identification plane, γ Is a constant, N is the total number of contents. Further, t _i is a determination reliability of the content i, and T is a reference value (constant) of the determination reliability. As a result, the error can be weighted with the determination reliability. Therefore, an error is less likely to be allowed for a content with a high determination reliability, and an error is easily allowed for a content with a low determination reliability.

  A schematic diagram of a separation plane generated using this extended SVM method is shown in FIG. In FIG. 8, content A is content with low reliability for satisfaction determination, and content B is content with high reliability for satisfaction determination. As a result, although the content A is determined to be unsatisfactory content, the reliability is low, and therefore the content A is allowed even outside the unsatisfactory class. That is, for content A, an error is allowed.

  Further, the content B is determined to be a satisfactory content with high satisfaction determination reliability. Therefore, the separation plane w is generated so as to satisfy these.

Next, another SVM method for excluding contents with relatively low determination reliability will be described. For example, regarding the content A and the content B shown in FIGS. 7 and 8, it is determined that the content A has a low determination reliability (below a predetermined value) and the content B has a high determination reliability (a predetermined value or more). When the reliability is low, processing that does not take into account when generating the separation plane can be considered. For example, the separation plane can be generated by using the following formula (3).

Here, 1 / ‖w‖ is a distance between the identification plane and the separation plane, ξ _i is a distance indicating how far the content i, which is sample data, has entered beyond the identification plane, and γ is a constant. N1 is the total number of contents whose determination reliability is equal to or higher than a predetermined value.

  As a result, the separation plane is generated only for content having a high determination reliability, that is, a predetermined value or more, and it can be expected to generate a separation plane with higher accuracy. In this method, N1 contents whose determination reliability is equal to or higher than a predetermined value (predetermined constant) are targeted. However, the present invention is not limited to this. For example, a predetermined coefficient α is added to the average value of the determination reliability. The content to be a target when the separation plane for N1 contents is generated by multiplying (> 0) as a predetermined value.

Furthermore, a method combining the methods represented by the equations (2) and (3) can be considered. That is, the separation plane can be generated by an expression represented by the following expression (4).

Again, as described above, 1 / ‖w‖ is the distance between the identification plane and the separation plane, ξ _i is the distance indicating how far the sample data content i has entered the opposite side beyond the identification plane, γ Is a constant, and N1 is the total number of contents whose content determination reliability is equal to or higher than a predetermined value. Further, t _i is a determination reliability of the content i, and T is a reference value (constant) of the determination reliability.

  This makes it possible to minimize the sum of products of data errors that meet the criteria for judgment reliability and the judgment reliability, and to avoid more accurate separation by not targeting content with a reliability lower than a predetermined value. It can be expected to generate a plane.

  In this way, the separation plane generation unit 109 separates the satisfactory content and the unsatisfactory content using the separation plane, and the content transmission unit 102 can transmit the content located on the satisfactory content side as recommendation information.

  Next, processing of the recommendation information distribution apparatus 100 will be described. FIG. 9 is a flowchart showing processing of the recommendation information distribution apparatus 100 when generating a separation plane. Each time a content request from a client terminal is received by the content request receiving unit 101 and the content is transmitted by the content transmitting unit 102, a content ID indicating the content browsed by the user is described in the browsing history table 104b. The page ID and the browsing date and time, and the content ID and the date and time of the content that has not been browsed are stored in the non-browsing history table 104d (S101).

  When the recommendation request is received by the recommendation request receiving unit 105, the page ID of the content browsed by the user for each content is acquired from the browsing history table 104b by the browsing level calculating unit 106 (S102). Then, the browsing level calculation unit 106 calculates the cumulative value for each content of the browsing level and the non-browsing level acquired from the browsing history table 104b, and the cumulative value of the calculated browsing level and the non-browsing level is the satisfaction determination table 104c. Are described in the browsing degree column and the non-browsing degree column (S103).

  The degree of interest is calculated by the determination unit 107 based on the browsing degree and the non-browsing degree stored in the satisfaction determination table 104c. Based on the degree of interest, the determination unit 107 determines whether the content is to be a satisfaction class or a dissatisfaction class for each content, and the determination result is described in the satisfaction determination table 104c (S104).

  Next, the reliability calculation unit 108 calculates the determination reliability of the satisfaction determination result (S105). Then, the separation plane generation unit 109 extracts a feature vector from the plurality of contents stored in the content storage unit 103b, and generates a separation plane for separating the satisfaction class and the dissatisfaction class using the feature vector. (S107). With respect to the method for generating the separation plane, as described above, the separation plane in consideration of the determination reliability is generated using the extended SVM method.

  Next, utilization of this separation plane will be described. FIG. 10 is a flowchart showing a transmission process of content separated into satisfaction classes using a separation plane. A content outline of content belonging to the satisfaction class classified based on the separation plane generated by the separation plane generation unit 109 among the contents as distribution candidates is extracted from the content list as recommendation information by the content transmission unit 102 ( S201). Then, the recommendation information extracted by the content transmission unit 102 is transmitted to the client terminal (S202). In this way, recommendation information can be distributed using the separation plane.

  Next, the operation and effect of the recommendation information distribution apparatus 100 configured as described above will be described. In the recommendation information distribution apparatus 100 of the present embodiment, the content transmission unit 102 extracts a content list indicating a plurality of contents including a page ID that is a score indicating the degree of interest from the content list storage unit 103a and transmits the content list to the client terminal. When one content is selected from the content list at the client terminal, the one content is transmitted to the client terminal. Then, the content transmitting unit 102 derives the browsing score based on the page ID included in the transmitted one content, and describes the derived score in the browsing history table 104b. Along with this processing, the content other than one content displayed in the content list is calculated as the non-viewing degree because it has not been browsed and described in the browsing history table 104b. In this way, the browsing history table 104b is updated with browsing history information every time content is transmitted.

  When the recommendation request receiving unit 105 receives the recommendation request, the browsing degree calculating unit 106 calculates the cumulative value of the browsing degree and the non-browsing degree stored in the browsing history table 104b for each content, and the determining unit 107 Determines the interest level of each content based on the accumulated value. For example, if the interest level obtained by subtracting the non-viewing level from the browsing level is a positive value, it is determined that the content is satisfactory.

  Then, the reliability calculation unit 108 calculates the reliability of the determination result from the absolute value of the interest level, and the separation plane generation unit 109 sets a plurality of contents as a satisfaction class and dissatisfaction based on the calculated reliability. Generate a separation plane that separates into classes. For example, the separation plane is generated using the SVM method. The content transmission unit 102 transmits the content evaluated using the generated separation plane as recommendation information.

  As a result, unknown content can be separated into a satisfaction class and a dissatisfaction class using a separation plane, and content satisfied by the user can be provided as recommendation information. In particular, by using the reliability of the satisfaction determination result, an accurate separation plane can be generated, and recommendation information with a high degree of interest for the user can be provided.

  Further, in the recommendation information distribution apparatus 100, the browsing degree calculation unit 106 does not browse unknown content similar to content that is selected in the content list but is not selected in the SAI that generates the satisfaction determination table 104c. Increase the degree. The determination unit 107 determines the interest level of each content using the increased non-browsing degree. As a result, the non-browsing degree can be propagated to similar contents, and more accurate reliability can be calculated with a small number of samples.

  Further, in the recommended information distribution apparatus 100, a separation plane is generated so as not to allow an error for each content according to the calculated degree of reliability of each content, so that the reliability is higher than that of highly reliable content. It is possible to generate an accurate separation plane according to reliability by allowing an error with respect to content having a low degree. That is, even if it is determined that content with low reliability is satisfactory, for example, it is not clear whether it is really satisfactory or not. It is possible to generate a separation plane with high accuracy as a result of not giving importance.

DESCRIPTION OF SYMBOLS 100 ... Recommendation information delivery apparatus, 101 ... Content request receiving part, 102 ... Content transmission part, 103 ... Storage part, 103a ... Content list storage part, 103b ... Content storage part, 103c ... Score table according to page ID, 104a ... Content characteristic Vector management table 104b ... browsing history table 104c ... satisfaction determination table 104d ... non-browsing history table 105 ... recommend request receiving unit 106 ... browsing degree calculation unit 107 ... determination unit 108 ... reliability calculation unit 109 ... separation plane generation unit, 110 ... content evaluation unit.

Claims (5)

Transmitting a content list showing a plurality of contents to a client terminal, and when one content is selected from the content list in the client terminal, content transmitting means for transmitting the one content;
The browsing degree for one content transmitted by the content transmitting means and the non-viewing degree for other contents other than the one content displayed in the content list are stored, and the content is transmitted by the content transmitting means. Browsing history storage means for storing the browsing degree and the non-browsing degree each time;
Determination means for determining the degree of interest of each content based on the browsing degree and non-browsing degree of each content stored in the browsing history storage means;
Reliability calculation means for calculating the reliability of the determination result of each content determined by the determination means based on the browsing degree and non-viewing degree of each content stored in the browsing history storage means;
Separation plane generation means for generating a separation plane for separating each content into a satisfaction class and a dissatisfaction class based on the reliability of each content calculated by the reliability calculation means;
Recommend transmission means for transmitting content evaluated using the separation plane generated by the separation plane generation means as recommendation information;
A recommendation information distribution apparatus comprising: A browsing degree calculating means for increasing a non-browsing degree with respect to an unknown content that is similar to the content that is selected in the content list but is not selected;
The recommendation information distribution apparatus according to claim 1, wherein the determination unit determines an interest level of each content using the non-browsing degree increased by the browsing degree calculation unit.   2. The separation plane generation unit generates the separation plane according to the degree of reliability of each content calculated by the reliability calculation unit so as not to allow an error with respect to each content. Or the recommendation information delivery apparatus of 2.   The said separation plane production | generation means produces | generates a separation plane using the content by which the reliability calculated by the said reliability calculation means to a predetermined order | rank, It is any one of Claim 1 to 3 characterized by the above-mentioned. Recommendation information distribution device. Transmitting a content list indicating a plurality of contents to a client terminal, and when one content is selected from the content list in the client terminal, a content transmission step of transmitting the one content;
The browsing degree for one content transmitted in the content transmission step and the non-viewing degree for other content other than the one content displayed in the content list are stored, and the content is transmitted in the content transmission step. A browsing history storage step for storing the browsing degree and the non-browsing degree in the browsing history storage means each time;
A determination step of determining an interest level of each content based on a browsing level and a non-browsing level of each content stored in the browsing history storage unit;
A reliability calculation step of calculating the reliability of the determination result of each content determined by the determination step based on the browsing level and the non-browsing level of each content stored in the browsing history storage unit;
A separation plane generation step for generating a separation plane for separating each content into a satisfaction class and a dissatisfaction class based on the reliability of each content calculated in the reliability calculation step;
A recommended transmission step of transmitting content evaluated using the separation plane generated by the separation plane generation step as recommendation information;
A recommendation information delivery method comprising:

Images