JP6167029B2 - RECOMMENDATION INFORMATION GENERATION DEVICE AND RECOMMENDATION INFORMATION GENERATION METHOD - Google Patents

Description

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

  A mail order site introduces a product based on the purchase history of other users. In such a site, recommendation information for introducing a product related to the product is generated by batch processing with reference to the purchase history of other users in advance for all the products. Generally, by calculating the inner product of content feature vectors, the similarity is grasped, and recommendation information is generated so that the contents are arranged in descending order of similarity.

  However, generally, there are a large number of products and the like, and such batch processing takes time. For example, when there are 1 million contents, it is necessary to perform 1 million × 1 million arithmetic processing, and the processing becomes enormous.

  In addition, in general digital content other than mail order sites, such as news articles, it is necessary to constantly update content provided as related content with an early expiration date. In general, digital content has a large long tail and is not seen by anyone. Therefore, calculation for such content in advance is a waste of calculation resources.

  Here, when providing the recommendation information, it is considered that the recommendation information is provided separately in a batch processing portion and a real-time processing portion. The following Patent Document 1 discloses a technique that generates recommendation information by batch processing and real-time processing.

JP 2012-216175 A

  However, in the technique described in Patent Document 1, it is intended to generate recommendation information based on the correlation between users at the time of batch processing, and generates recommendations for users who are not scheduled to access. It is possible to perform a complicated calculation process. Thus, there is still a waste of computing resources.

  Therefore, in order to solve the above-described problems, the present invention provides a recommendation information generation device and a recommendation information generation method that can generate recommendation information without spending unnecessary calculation resources in performing batch processing. For the purpose.

  In order to solve the above-described problem, the recommendation information generation apparatus according to the present invention supports one or a plurality of contents for each feature element constituting the content feature vector based on the content feature vector of the content to be recommended. An inverted index table generating means for generating an attached inverted index table, an inverted index storage means for storing the inverted index table generated by the inverted index table generating means, and when a recommendation request is received from a user terminal, the inverted index Recommendation information generating means for generating recommendation information in accordance with the display order of contents at the user terminal determined based on the transposed index table stored in the storage means; and the recommendation information generating means Transmitting means for transmitting the recommended information, wherein the transposed index table generating means executes a transposed index table generating process when a predetermined condition is satisfied, and the recommended information generating means is a user Each time a recommendation request from a terminal is received, display order determination processing is executed.

  Also, the recommendation information generation method of the present invention is a recommendation information generation method in the recommendation information generation apparatus, wherein one or a plurality of feature information constituting the content feature vector is provided for each feature element based on the content feature vector of the content to be recommended. A transposed index table generating step for generating a transposed index table associated with the content, a storing step for storing the transposed index table generated by the transposed index table generating step in a transposed index storage unit, and a recommendation request from the user terminal. When accepted, it generates recommendation information according to the display order of the content on the user terminal determined based on the transposed index table stored in the transposed index storage means. A recommended information generating step, and a transmitting step for transmitting the recommended information generated by the recommended information generating step. The transposed index table generating step generates a transposed index table when a predetermined condition is satisfied. A process is executed, and the recommendation information generation step executes a display order determination process each time a recommendation request from a user terminal is received.

  According to this invention, based on the content feature vector of the content to be recommended, a transposed index table in which one or a plurality of contents are associated with each feature element constituting the content feature vector is generated, and this transposed index is generated. Remember the table. Then, when a recommendation request is received from the user terminal, recommendation information is generated according to the display order of contents in the user terminal determined based on the transposed index table, and the recommendation information is transmitted. Thus, the transposed index table generation process is executed when a predetermined condition is satisfied, and the display order determination process is executed every time a recommendation request from the user terminal is received. Therefore, it is possible to provide recommendation information even for a service that requires a high-speed update process, by ending the process that requires calculation time in advance. Even if only some contents are intensively targeted for recommendation, calculation is not performed for all the contents, so that the calculation resource is not wasted. .

  Further, in the recommendation information generating apparatus of the present invention, the recommendation information generating means aggregates, for each content, a duplication number that is the number of contents that are described in duplicate by changing the feature elements based on the transposed index table. And a generating unit that generates a management list in which the duplication numbers totaled by the totaling unit are associated with arbitrary feature elements among the characteristic elements constituting the content feature vector of the content. Extraction means for generating recommendation information by extracting the content so that the content can be displayed in the order of multiple duplications on the user terminal based on the generated management list.

  According to this invention, based on the transposed index table, the duplication number, which is the number of contents described in duplicate by changing the characteristic elements, is tabulated for each content, and the tabulated duplication number of the content is counted. A management list associated with an arbitrary feature element among the feature elements constituting the content feature vector is generated, and content can be displayed in the order of multiple overlaps on the user terminal based on the generated management list Recommendation information is generated by extracting content. Since simple arithmetic processing such as counting the number of overlaps is employed, it is suitable for real-time processing. Therefore, by generating an inverted index table in advance, recommendation information can be generated by real-time processing.

  Further, the recommendation information generating apparatus according to the present invention includes a content feature vector generating unit that generates a content feature vector in which content features are expressed based on predetermined feature elements, and the content feature vector generating unit. Content feature vector storage means for storing a content feature vector for each content, and the transposed index table generating means uses the content feature vector stored in the content feature vector storage means to generate a transposed index table. Generate.

  According to the present invention, a content feature vector in which content features are expressed based on predetermined feature elements is generated, and the generated content feature vector is stored for each content. Then, a transposed index table is generated using the stored content feature vector. As a result, it is possible to provide recommendation information even for services that require high-speed update processing by ending processing that takes a long time in advance.

  In the recommendation information generating apparatus according to the present invention, the transposed index table generating unit may perform batch processing on a predetermined number of content feature vectors when the number of content feature vectors stored in the content feature vector storage unit reaches a predetermined number. And execute together.

  According to the present invention, since the processing up to the generation of the inverted index table is performed by batch processing such as performing on the content feature vectors that have been collected in advance, the generation efficiency of the inverted index table can be improved. As a result, computational resources can be used efficiently.

  According to the present invention, it is possible to provide recommendation information even for a service that requires high-speed update processing, by ending processing that takes a long time in advance. Even if only some contents are intensively targeted for recommendation, calculation is not performed for all the contents, so that the calculation resource is not wasted. .

1 is a system configuration diagram using an information distribution server 100. FIG. 2 is a block diagram showing a functional configuration of an information distribution server 100. FIG. 2 is a hardware configuration diagram of an information distribution server 100. FIG. 5 is an explanatory diagram illustrating a specific example of a feature dictionary stored in a feature dictionary storage unit 107. FIG. It is explanatory drawing which shows metadata and a content feature vector. It is explanatory drawing which shows an evaluation value file. It is a specific example of an inverted index table. It is explanatory drawing of a content ID list. It is explanatory drawing of an intermediate | middle list. It is explanatory drawing which shows the production | generation process from the intermediate | middle list to the last list, and the production | generation process of the list of the recommendation content which is a final result. 4 is a block diagram illustrating functions of the mobile terminal 200. FIG. 4 is a flowchart showing processing of the information distribution server 100.

  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 system configuration diagram using an information distribution server 100 which is a recommendation information generation apparatus according to the present embodiment. As shown in FIG. 1, the information distribution server 100 can communicate with the mobile terminal 200 via a network and transmit content in accordance with user preferences of the mobile terminal 200. The mobile terminal 200 is a mobile phone or a smartphone, but is not limited to this, and may be a personal computer or a tablet terminal.

  FIG. 2 is a block diagram showing a functional configuration of the information distribution server 100 of the present embodiment. The information distribution server 100 includes a content feature vector generation unit 101 (content feature vector generation unit), a transposed index generation unit 102 (transposition index table generation unit), a content feature vector acquisition unit 103, a recommended content acquisition unit 104 (recommendation information generation). Means, aggregation means, generation means, extraction means), recommended content transmission section 105 (transmission means), metadata storage section 106, feature dictionary storage section 107, content feature vector storage section 108 (content feature vector storage means), and transposition. The index storage unit 109 (transposed index storage means) is included.

  FIG. 3 is a hardware configuration diagram of the information distribution server 100. As shown in FIG. 3, the information distribution server 100 shown in FIG. 2 physically includes one or more CPUs 11, a RAM 12 and a ROM 13 that are main storage devices, and an input device 14 such as a keyboard and a mouse that are input devices. The computer system includes an output device 15 such as a display, a communication module 16 that is a data transmission / reception device such as a network card, and an auxiliary storage device 17 such as a hard disk or a semiconductor memory. Each function in FIG. 2 operates the input device 14, the output device 15, and the communication module 16 under the control of the CPU 11 by reading predetermined computer software on the hardware such as the CPU 11 and the RAM 12 shown in FIG. 3. In addition, it is realized by reading and writing data in the RAM 12 and the auxiliary storage device 17. Hereinafter, each functional block will be described based on the functional blocks shown in FIG.

  The content feature vector generation unit 101 is a part that generates content feature vectors based on the metadata of each content stored in the metadata storage unit 106 and the feature dictionary stored in the feature dictionary storage unit 107. The generated content feature vector is stored in the content feature vector storage unit 108. The content feature vector generation unit 101 generates a content feature vector based on a feature dictionary when content registration is performed by a content provider, that is, when content and its metadata are received from the content provider. become.

  Here, generation of content feature vectors will be described with reference to the drawings. FIG. 4 is an explanatory diagram showing a specific example of the feature dictionary stored in the feature dictionary storage unit 107. This feature dictionary is composed of feature elements, matching targets, and matching information. Here, the feature element represents each element when expressed by a feature vector, and includes an ID and contents representing the feature. In FIG. 4, ID: 1 is assigned to “major” which is the content representing the feature element. The matching target defines information in the content to be matched when expressing the feature of the content, and is information indicating which part of the content is the matching target of the matching information. In FIG. 4, “description” is described as a matching target of the feature element “major”, and it is determined whether or not the word “major” is described in the “description” column of the metadata. It will be.

  Further, the matching information indicates a keyword to be matched and a value indicating the importance between the keyword and the feature element. This matching information is information for adjusting so-called notation fluctuations. For example, in the baseball terminology, “major” may be referred to as “major league”, but has substantially the same meaning. By associating such words indicating the same meaning in the matching information and slightly changing the importance value, the notation fluctuation can be reflected in the vector expression. In FIG. 4, the value “1” is associated with “major”, but the value “0.5” is associated with “major league”. As can be seen from this, for the word that is not the feature element itself, the expression in the content feature vector is weakened by lowering the importance value.

  Next, content metadata will be described. FIG. 5A is an explanatory diagram showing content metadata, and FIG. 5B and FIG. 5C are explanatory diagrams showing examples of content feature vectors generated based on the metadata. .

  As shown in FIG. 5A, the metadata includes a content ID, a category, and an explanatory text. Here, the content ID is an identification number for identifying the content, the category indicates the type of content, and the explanatory text indicates the outline of the content.

  The content feature vector generation unit 101 generates the content feature vector shown in FIG. 5B using the metadata and the feature dictionary. In FIG. 5B, numerical values constituting the feature vector are described for each content ID. Here, a content feature vector having a major, real estate, drama, etc. as a feature element is generated. .

  FIG. 5C shows a content feature vector normalized with respect to the content feature vector shown in FIG. As the normalization process, a general process can be used. The content feature vector generation unit 101 generates a content feature vector obtained by further normalizing the generated content feature vector, and stores the content feature vector in the content feature vector storage unit 108. The normalization process is not an essential process and may be omitted.

  The transposed index generation unit 102 is a part that generates a transposed index table in which content is associated with each feature element from any or all content feature vectors stored in the content feature vector storage unit 108. Here, the content feature vector is composed of feature elements having hundreds to thousands of dimensions. In the case of each content, when the property indicated by each feature element is expressed, a numerical value greater than 0 is described in each feature element. The above-described transposed index generation unit 102 generates a transposed index table in which the content ID of content in which a numerical value greater than 0 is described for each feature element is associated with each feature element, and this is stored in the transposed index storage unit 109. Remember.

  A method for generating an inverted index table will be described with reference to the drawings. First, an evaluation value file used for generating an inverted index table will be described prior to the generation method. FIG. 6 is an explanatory diagram showing an evaluation value file. As shown in FIG. 6, this evaluation value file associates a feature ID, a content ID, and an evaluation value. This evaluation value file indicates an evaluation value for the combination of the feature ID and the content ID, and is used when determining the priority order of the content in the transposed index table. For example, when a content ID is associated with a certain feature ID, it is used to determine the priority order of the content indicated by the content ID based on the evaluation value. This evaluation value file is created in advance by the operator of the recommendation service based on information from the content provider.

  Then, the transposed index generation unit 102 generates a transposed index table by associating contents for each feature element. That is, the transposed index generation unit 102 extracts, for each feature ID, the content ID of the content feature vector in which the feature element indicated by the feature ID is set to a value greater than 0, and this is indicated in the evaluation value file. Corresponding processing is performed in the order of evaluation values.

  FIG. 7 is a specific example of an inverted index table. As shown in FIG. 7, the transposed index table associates content IDs of one or more contents with each feature element that identifies the feature element. In the content ID column, a plurality of content IDs are described, which are described in the order based on the evaluation values indicated in the evaluation value file. For example, in the evaluation value file, when the evaluation value of the content ID: 11 corresponding to the feature ID: 1 is 0.25 and the evaluation value of the content ID: 21 corresponding to the feature ID: 1 is 0.20, the content Since the evaluation value of ID: 11 is higher, it is described before the content ID: 21.

  The inverted index generation unit 102 generates an inverted index table through such processing. The generated inverted index table is stored in the inverted index storage unit 109. Note that the inverted index generation unit 102 generates the inverted index table in advance by batch processing. For example, when the content feature vector stored in the content feature vector storage unit 108 reaches a predetermined number or when a predetermined condition such as once a month is satisfied, a transposed index table is generated, and the transposed index storage unit 109 Update the inverted index table.

  Next, each configuration in FIG. 2 will be described. When a recommendation request (that is, a search request) is made from the user, the content feature vector acquisition unit 103 generates a recommended content corresponding to the recommendation request based on a predetermined rule from the content feature vector storage unit 108. This is a part for acquiring the content feature vector of the determined content. For example, when browsing content in a certain site, in order to generate recommended content based on the content, a content feature vector of the content is acquired. Although one content feature vector is acquired here, for example, content feature vectors of one or a plurality of contents similar to the content may be acquired together. Similar determinations can be made, for example, by taking the inner product of content feature vectors.

  The recommended content acquisition unit 104 extracts a feature ID that is an identification number of a feature element based on the content feature vector acquired by the content feature vector acquisition unit 103. As a method of extracting the feature ID, a feature ID in which a numerical value greater than 0 is described in the content feature vector is extracted. Then, using this feature ID as a key, one or more content IDs corresponding to the feature ID are extracted in order to acquire recommendation content from the transposed index storage unit 109. That is, the content ID associated with the extracted feature ID is further extracted to generate a content ID list in which the feature ID and the content ID are associated with each other.

  Further, the recommended content acquisition unit 104 sequentially extracts one or more contents to be displayed on the mobile terminal 200 from the generated content ID list. That is, the recommended content acquisition unit 104 extracts content from the content ID list in an order according to a predetermined rule.

  This extraction process will be described in detail below. First, the recommended content acquisition unit 104 generates a list of content IDs corresponding to each feature ID based on the transposed index table, as shown in FIG.

  FIG. 8 is an explanatory diagram of a content ID list indicating the content ID corresponding to each feature ID generated based on the transposed index table. As described above, this content ID list is generated based on the characteristic elements of the content that is the source of the recommendation (for example, the browsed content). In FIG. As feature IDs, three featureId_a to featureId_c are extracted, and a list is formed in that order. Here, for the sake of convenience, it is assumed that the feature ID: featureId_a has a higher importance level, and the importance level is set lower as the feature ID_b and featureId_c become. This importance is associated with each feature ID.

  In FIG. 8, content IDs: contentsId_2, contentsId_5, and contentsId_6 are associated with feature ID: featureId_a. Similarly, the content ID: featuresId_b is associated with content IDs: contentsId_3, contentsId_5, and contentsId_1, and the feature ID: featureId_c is associated with content IDs: contentsId_5, contentsId_1, and contentsId_4. Each content ID is extracted based on the transposed index table.

  Next, when the same content ID appears in the content ID list in FIG. 8, the recommended content acquisition unit 104 counts the number as the overlapping number. Then, an intermediate list is generated by adding duplicate columns to the content ID list.

  FIG. 9 is an explanatory diagram of a specific example of the intermediate list. This intermediate list is a management list for managing feature IDs and content IDs, and stores feature IDs, content IDs, and overlapping numbers in association with each other.

  In FIG. 9, the overlap number of content ID: contentsId_5 is 3. Here, since the content ID: contentsId_5 appears in each of the feature IDs: featureId_a to featureId_c, the overlap number can be calculated as 3 by counting each.

  Then, the recommended content acquisition unit 104 further processes this intermediate list to generate a final list. FIG. 10 shows a generation process from the intermediate list to the final list, and a process of generating a list of recommended contents as the final result.

  FIG. 10A is an explanatory diagram showing that the same content ID is deleted from the intermediate list. As shown in the figure, since the content ID: contentsId_5 appears in any of the feature IDs: featureId_a, featureId_b, and featureId_c, 3 is associated with only the content ID: contentuId_5 of the upper feature ID: featureId_a as a duplication number. The field of feature ID: featureId_b and featureId_c is deleted. In FIG. 10A, the deletion state is indicated by a strikethrough.

  FIG. 10B is the final list of the content ID list that has been deleted. As is apparent from the figure, the content IDs: contentsId_5 and contentsId_1 are deleted from the correspondence with the feature ID: featureId_b and the feature ID: featureId_c, respectively.

  And the recommendation content acquisition part 104 determines the display order in the portable terminal 200 by selecting the content as recommendation content in order based on the last list of the content ID list | wrist of FIG.10 (b). That is, in this final list, the content IDs are arranged in order from the top of the feature ID, and further, in the same feature ID column, the content IDs are arranged in the descending order of the overlapping number. The recommended content acquisition unit 104 can determine the display order by selecting a predetermined upper number of content IDs.

  In addition to the above, the content extraction method from the final list may prioritize multiple content IDs. That is. The content IDs may be extracted in descending order of the number of duplicates, and then the higher-ranked feature IDs may be extracted in order.

  In FIG. 10B, the overlapping number of the content ID: contentsId_5 corresponding to the topmost feature ID: featureId_a is the largest, and the next ones are the content IDs: contentsId_2 and contentsId_6. The plurality is 1.

  In the next order of the feature ID: featureId_a is the feature ID: featureId_b, and in the same feature ID column, the content IDs: contentsId_1 and contentsId_3 are arranged in the descending order. The same applies to the feature ID of the next rank.

  The recommended content acquisition unit 104 can determine the display order as the recommended content by sequentially selecting the upper predetermined number of the contents arranged in the order of the feature ID and the overlapping order. In the present embodiment, as shown in FIG. 10C, content IDs for the top five items are selected. Then, the recommended content acquisition unit 104 acquires the content corresponding to the content ID selected in this way as recommended content in the display order.

  The recommended content transmission unit 105 is a part that sequentially transmits the content extracted by the recommended content acquisition unit 104 to the mobile terminal 200 as recommended content.

  Next, the portable terminal 200 that makes a request for recommended content to the information distribution server 100 will be described. FIG. 11 is a block diagram illustrating functions of the mobile terminal 200. As shown in FIG. 11, the mobile terminal 200 includes a browser 201, a history transmission unit 202, a recommendation request transmission unit 203, and a distribution information display unit 204. The portable terminal 200 is realized by the hardware configuration shown in FIG. 3, and is realized by a computer system including a CPU, a RAM, a ROM, and the like. Hereinafter, each component will be described.

  The browser 201 is a part for browsing various contents via the Internet or the like. Generally, it is configured by an application.

  The history transmission unit 202 is a part that transmits history information including a search request or a browsing request executed by the browser 201 to the outside.

  The recommendation request transmission unit 203 is a part that transmits a recommendation request to the information distribution server 100. The recommendation request itself may be transmitted, or when a certain content is browsed in the browser 201, a recommendation request related to the content may be transmitted.

  The distribution information display unit 204 is a part that receives and displays recommendation information for the recommendation request transmitted by the recommendation request transmission unit 203. Here, the recommended content that has been processed for the display order of the content as described above is displayed in the display order.

  Next, processing of the information distribution server 100 and the mobile terminal 200 configured as described above will be described. FIG. 12 is a flowchart showing processing of the information distribution server 100.

  In the information distribution server 100, when content is registered by the content provider (S101), the content feature vector generation unit 101 generates a content feature vector (S102). Then, the generated content feature vector is stored in the content feature vector storage unit 108 (S103).

  Next, a transposed index table is generated by the transposed index generation unit 102 (S104), and the transposed index table is stored in the transposed index storage unit 109 (S105). The process of S104 may be a series of processes from S103, or the transposed index table may be generated after content feature vectors are accumulated to some extent.

  The processing from S101 to S105 is performed before the recommendation content generation processing, and is executed by so-called batch processing.

  Next, recommended content is generated based on the transposed index table generated in this way.

  First, when a recommended content request is received from the mobile terminal 200 (S201), the content feature vector acquisition unit 103 acquires one or more content feature vectors from the content feature vector storage unit 108 (S202). Then, the recommended content acquisition unit 104 generates a content ID list based on the transposed index table (S203). Here, the content ID list shown in FIG. 8 is acquired.

  Then, the recommended content acquisition unit 104 executes content extraction processing (S204). That is, from the content ID list of FIG. 8, a plurality of content IDs are obtained, and an intermediate list is generated (see FIG. 9). Next, a final list (see FIG. 10B), which is a content ID list arranged based on the overlap, is generated, and content as recommended content is extracted based on the final list (FIG. 10). (See (c)).

  The extracted content is transmitted as recommended content to the mobile terminal 200 by the recommended content transmitting unit 105 (S205).

  In this way, an inverted index table is generated by batch processing, and thereafter, every time a recommendation request is received, recommendation processing using the inverted index table can be performed by real-time processing.

  Next, operational effects of the information distribution server 100 of the present embodiment will be described. According to the information distribution server 100 of the present embodiment, the content feature vector generation unit 101 generates a content feature vector in which content features are expressed based on predetermined feature elements, and the content feature vector storage unit 108 The generated content feature vector is stored for each content. Then, the transposed index generation unit 102 selects, for each feature element constituting the content feature vector stored in the content feature vector storage unit 108, the content characterized by the feature element, and selects each selected content. A transposed index table associated with the feature element is generated. The transposed index storage unit 109 stores the generated transposed index table. The processing from the generation of the content feature vector to the generation of the transposed index table is terminated before the recommendation request from the portable terminal 200 is received.

  Next, when a recommendation request is received from a user terminal by a recommendation request receiving unit (not shown), the recommended content acquisition unit 104 is described in duplicate from the transposed index table stored in the transposed index storage unit 109. The number of contents that have been added is aggregated for each content.

  Then, the recommended content acquisition unit 104 associates the total number of contents with an arbitrary feature element (in this embodiment, the highest-order feature element) among the feature elements constituting the content feature vector of the content. Is generated. And the recommendation content acquisition part 104 produces | generates a final list from this intermediate list, and extracts a content so that a content can be displayed in order with many duplication in a user terminal. The recommended content transmission unit 105 transmits the content so that the extracted content is displayed on the mobile terminal 200.

  As a result, processing that takes a long time can be completed in advance, and can be used for services that require high-speed update processing. Even if only a part of the content is intensively targeted for recommendation, the calculation resource is not wasted because it is not calculated for all the content. .

  Further, in the information distribution server 100 of the present embodiment, the transposed index generation unit 102 collects the predetermined number of content by batch processing when the content feature vectors stored in the content feature vector storage unit 108 reach the predetermined number. May be executed.

  As a result, the processing up to the generation of the inverted index table is executed by batch processing such as performing on the content feature vectors collected in advance, so that the generation efficiency of the inverted index table can be improved. In addition, computational resources can be used efficiently.

  Note that, as conditions other than those described above, the transposed index table may be generated in accordance with a predetermined cycle, a predetermined time or a predetermined date and time, or according to an instruction of an operator who provides this recommendation service.

DESCRIPTION OF SYMBOLS 100 ... Information delivery server, 101 ... Content feature vector production | generation part, 102 ... Transposition index production | generation part, 103 ... Content feature vector acquisition part, 104 ... Recommended content acquisition part, 105 ... Recommended content transmission part, 106 ... Metadata storage part, DESCRIPTION OF SYMBOLS 107 ... Feature dictionary memory | storage part, 108 ... Content feature vector memory | storage part, 109 ... Transposition index memory | storage part, 200 ... Mobile terminal, 201 ... Browser, 202 ... History transmission part, 203 ... Recommendation request transmission part, 204 ... Distribution information display part .

Claims (5)

Based on the content feature vector of the content to be recommended, a transposed index table generating unit that generates a transposed index table in which one or more contents are associated with each feature element constituting the content feature vector;
A transposed index storage unit for storing the transposed index table generated by the transposed index table generating unit;
When a recommendation request is received from a user terminal, recommendation information generating means for generating recommendation information according to the display order of contents in the user terminal determined based on the transposed index table stored in the transposed index storage means;
Transmitting means for transmitting the recommendation information generated by the recommendation information generating means;
With
The transposed index table generating means executes a transposed index table generating process when a predetermined condition is satisfied,
The recommendation information generation unit is a recommendation information generation device that executes a display order determination process each time a recommendation request from a user terminal is received. The recommendation information generation means includes:
Based on the transposed index table, a totaling unit that counts the duplication number, which is the number of contents that are duplicated by changing the characteristic elements, for each content;
Generating means for generating a management list in which the duplication numbers totaled by the totaling means are associated with arbitrary feature elements among the feature elements constituting the content feature vector of the content;
Based on the management list generated by the generating means, extracting means for generating recommendation information by extracting the content so that the content can be displayed in the order of multiple duplication at the user terminal;
The recommendation information generating apparatus according to claim 1, comprising: Content feature vector generation means for generating a content feature vector in which the feature of the content is expressed based on a predetermined feature element;
Content feature vector storage means for storing the content feature vector generated by the content feature vector generation means for each content;
With
The recommendation information generating apparatus according to claim 1, wherein the transposed index table generating unit generates a transposed index table using a content feature vector stored in the content feature vector storage unit.   The recommendation information generating apparatus according to claim 3, wherein the transposed index table generating unit collectively executes the batch processing when the number of content feature vectors stored in the content feature vector storage unit reaches a predetermined number. . In the recommendation information generation method in the recommendation information generation device,
Based on the content feature vector of the content to be recommended, a transposed index table generating step for generating a transposed index table in which one or a plurality of contents are associated with each feature element constituting the content feature vector;
A storage step of storing the inverted index table generated by the inverted index table generating step in an inverted index storage means;
When a recommendation request is received from a user terminal, a recommendation information generation step of generating recommendation information according to the display order of content in the user terminal determined based on the transposed index table stored in the transposed index storage unit;
A transmission step of transmitting the recommendation information generated by the recommendation information generation step;
With
The transposed index table generating step performs a transposed index table generating process when a predetermined condition is satisfied,
The recommendation information generation step is a recommendation information generation method for executing a display order determination process each time a recommendation request from a user terminal is received.

Images