JP6433270B2 - Content search result providing system and content search result providing method - Google Patents

Description

  The present invention relates to a content search result providing system and a content search result providing method.

  2. Description of the Related Art Conventionally, there is a technique for outputting a search result corresponding to a search query based on a user's past search history in order to provide content according to the user's interests. For example, in the technique described in Patent Document 1, the result of analyzing the content of a web page requested to be browsed by a user's click operation is stored in a database, and the priority of search results to be output is changed based on the database.

JP 2011-227572 A

  In the above prior art, the user's palatability is reflected by using the search click history (hereinafter referred to as “search click history”) indicating that the content is selected from the search result by the user's click operation or the like. Provide content. Therefore, when there is not enough search click history of a user who makes a search request, it is not possible to provide content that reflects the user's preference.

  Therefore, the present invention provides a content search result providing apparatus and a content search result providing method capable of providing content reflecting the user's preference even when there is not enough search click history of the user who makes the search request. The purpose is to provide.

  The content search result providing system according to the present invention includes a query category dictionary in which information relating a predetermined search query and a degree of correspondence to a predetermined search query for each content category is stored, and a past transmitted from a user terminal Using the search history acquisition means to acquire the search query history indicating the search query of the search query and the search query history, the information that correlates the degree of correspondence with the past search query for each content category is extracted from the query category dictionary and extracted. User feature amount calculating means for calculating a user feature amount indicating the degree of correspondence of each user's preference with respect to a content category, and a category corresponding to a search query at the time of a search request transmitted from a user terminal. Search results for content that contains information indicating content associated with Comprises a search result acquisition means for, based on the user's feature quantity, and output means for outputting the search results of the content, the.

  Also, the content search result providing method according to the present invention provides a content search result provided with a query category dictionary in which information relating a predetermined search query and a degree of correspondence to the predetermined search query for each category of content is stored. A content search result providing method, which is an operation method of the system, including a search history acquisition step for acquiring a search query history indicating a past search query transmitted from a user terminal, and a query category dictionary using the search query history Extracting information that associates the degree of correspondence to past search queries for each content category, and calculating a user feature amount that indicates the degree of preference for each user with respect to the content category based on the extracted information Amount calculation step and search at the time of search request sent from user terminal Including a search result obtaining step of obtaining search results of content including information indicating the content associated with the category with corresponding area, based on the user's feature quantity, and outputting the search results of content and.

  According to such a content search result providing system and content search result providing method, information associating the degree of correspondence with a past search query for each content category is extracted from the query category dictionary, and based on the extracted information. Then, the user feature amount indicating the degree of correspondence of the preference for each user with respect to the content category is calculated. Therefore, the user feature amount is calculated without using the search click history of the user who makes the search request. Since the search result of the content corresponding to the search query at the time of the search request is output based on the user feature amount, the user's preference is reflected even when there is no search click history of the user who makes the search request. Search results can be output. As described above, even when there is no search click history of a user who makes a search request, it is possible to provide content that reflects the user's preference.

  The content search result providing system according to the present invention further comprises rank calculating means for determining the order of information indicating the content included in the content search result based on the user feature amount, and the output means is provided by the rank calculating means. Content search results may be output according to the determined order. In this case, since the search result of the content in the order based on the user feature amount is output, even when there is no search click history of the user who makes the search request, the search result arranged in the order corresponding to the user's preference Can be output.

Further, in the content search result providing system according to the present invention, the predetermined search query includes a past search query, and the search history acquisition unit corresponds to the past search query and the past search query for each of a plurality of users. of the content, the user has to acquire a multi-user history is information associating the category of content selected, based on a plurality user history, further Ru comprising a creating means for creating a query category dictionary. In the content search result providing method according to the present invention, the predetermined search query includes a past search query, and in the search history acquisition step, the past search query and the content corresponding to the past search query for each of a plurality of users. A plurality of user histories that are information associated with the content category selected by the user, and a creation step of creating a query category dictionary based on the plurality of user histories. In this case, since a query category dictionary is created based on a plurality of user histories in which past search queries and categories are associated with a plurality of users, the reliability of information stored in the query category dictionary can be further increased. .

  The content search result providing system according to the present invention may further include search means for searching for content corresponding to the search query at the time of the search request transmitted from the user terminal. In this case, the content corresponding to the search query at the time of the search request transmitted from the user terminal can be searched by the search means.

  According to the present invention, there are provided a content search result providing system and a content search result providing method capable of providing content reflecting the user's preference even when there is no search click history of a user who makes a search request. can do.

It is a block diagram which shows the function structure of the content search result provision system which concerns on this embodiment. It is a hardware block diagram of an integrated server and a search server. It is explanatory drawing which shows the content of the information memorize | stored by the search history information storage part. It is explanatory drawing which shows the content of the information stored in a query category dictionary. It is explanatory drawing which shows the content of the information memorize | stored by the user feature vector memory | storage part. It is a table | surface which shows the calculation process of the user feature vector evaluation value when the search action of the 1st time is performed. It is a table | surface which shows the calculation process of the user characteristic vector evaluation value when the search action of the 2nd time is performed. It is explanatory drawing which shows the method of rearranging the list information of the search result content from a search result content acquisition part based on a user feature vector. It is a sequence diagram which shows the whole operation | movement procedure of the content search result provision system shown by FIG. It is a flowchart which shows the whole operation | movement procedure of the content search result provision system shown by FIG. It is a flowchart which shows the detailed processing method of S20 shown by FIG. It is a flowchart which shows the detailed processing method of S30 shown by FIG. It is a flowchart which shows the detailed processing method of S32 shown by FIG. It is a flowchart which shows the detailed processing method of S33 shown by FIG. It is a flowchart which shows the detailed processing method of S70 shown by FIG.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a block diagram showing a functional configuration of a content search result providing system according to the present embodiment. As shown in FIG. 1, the content search result providing system 200 includes an integrated server (integrated engine) 300 and a search server (search engine) 400. A user terminal 100 such as a mobile phone is connected to the integrated server 300 via a network N.

  For example, the integrated server 300 provides the user terminal 100 with a content distribution service operated by a telecommunications carrier via the network N that is a mobile communication network. The content distribution service is a service that distributes various contents in response to a request from a user, for example, on a network N of a telecommunications carrier. Content means the content or contents of information. The content includes digital content such as music, moving images, electronic books, and games. The content includes product information handled by an online shopping service or the like.

  The content distributed by the content distribution service is associated with metadata (attribute information) indicating in advance what kind of information the content is. For example, metadata such as a title or category is associated with the content in advance by a communication carrier. The title is a name given to the content. For example, when the content distributed in the content distribution service is a song, the song name is given as the title for the song. A category is a classification for classifying content according to its nature. For example, if the content distributed in the content distribution service is an idle song, the content has the property of music and is associated with the category “music”. Further, since the content also has the property of an idol song, it is also associated with the category “idol”.

  Note that, by associating a category with the content distributed by the content distribution service in advance, the content distribution service can distribute, for example, content organized by category. The content may include not only the content itself but also content metadata.

  The integrated server 300 receives a search query from the user terminal 100 and accepts a search request. A search query indicates a search request for extracting information as a search result. The search query is, for example, a word or a phrase (compound word) that is input when searching on the user terminal 100. For example, when a user inputs a word such as an artist name in the search window on the content distribution service displayed on the user terminal 100, the input word (search query) is accepted as a search request in the content distribution service. It is done. Thereby, the integrated server 300 receives a search request from the user terminal 100.

  The integration server 300 acquires the search result of the content corresponding to the search query from the search server 400. The content corresponding to the search query indicates content that hits (applies to) a search request by the search query, such as content related to a word or a phrase indicated by the search query. The search server 400 calculates and acquires a plurality of contents that hit the search request based on the search query. Hereinafter, a plurality of contents acquired as search results based on the search query will be simply referred to as “search result contents”. The integrated server 300 outputs information indicating the search result content acquired by the search server 400 to the user terminal 100 as a search result. The user terminal 100 includes a display unit such as a display, and displays information indicating the search result content output from the integrated server 300 on the display unit. In the display unit of the user terminal 100, the information indicating the search result content is list information including information indicating the content such as link information indicating the location of each content included in the search result content on the network. Hereinafter, the information indicating the search result content is also simply referred to as “search result content list information”.

  In the user terminal 100, information indicating the desired content is selected by the user from such search result content list information. Here, selection refers to selecting the content indicated by the information by selecting information indicating one content from the list information. The selection includes not only selection meaning browsing (for example, link selection of content detailed information) but also selection for purchase (for example, link selection for content purchase). The selection is realized by, for example, a click operation. Hereinafter, in the present embodiment, the case where the user selects content by a click operation will be described. Hereinafter, selection of content by a click operation is simply referred to as “click”. For example, when the user clicks on any of the search result content list information displayed on the display unit of the user terminal 100, it indicates that the content indicated by the clicked information has been selected. When content is selected in this way, a content transmission request is made from the user terminal 100 to the integrated server 300 so as to acquire the selected content.

  FIG. 2 is a hardware configuration diagram of the integrated server 300 and the search server 400. As shown in FIG. 2, the integrated server 300 and the search server 400 shown in FIG. 1 physically include one or a plurality of CPUs 11, a RAM 12 and a ROM 13 that are main storage devices, a keyboard and a mouse that are input devices, and the like. The computer system includes an input device 14, an output device 15 such as a display, a communication module 16 that is a data transmission / reception device such as a network card, an auxiliary storage device 17 such as a semiconductor memory, and the like. Each function of the content search result providing system is configured such that one or a plurality of predetermined computer software is read on the hardware such as the CPU 11 and the RAM 12 shown in FIG. This is realized by operating the output device 15 and the communication module 16 and reading and writing data in the RAM 12 and the auxiliary storage device 17.

  Hereinafter, the functional configuration of the integrated server 300 will be described with reference to FIG. 1 again. The integrated server 300 includes a search history acquisition unit 301 (search history acquisition unit), a search history information storage unit 302 (storage unit), a query category dictionary 303, a query category dictionary creation unit 304 (creation unit), and a user feature vector calculation unit ( (User feature amount calculation means) 305, user feature vector storage section 306, search request reception section 307, search request transmission section 308, search result content acquisition section 309 (search result acquisition means), rank calculation section 310 (rank calculation means), And the search result content output part 311 (output means) is comprised.

  The search history acquisition unit 301 is selected in the user terminal 100 among the user ID of the user terminal 100 that has made the search request, the search query from the user terminal 100 of the user ID, and the search result of the content based on the search query. Get content and. The user ID of the user terminal 100 is a number or abbreviation for identifying the user terminal 100. The user ID is information necessary for each user to use the content distribution service, for example. When the user terminal 100 uses the content distribution service, the user ID is input by the user on the login screen of the content distribution service displayed on the display unit of the user terminal 100. When the user ID is input, this user ID is stored in the user terminal 100.

  The search query is input by the user in the search window on the content distribution service displayed on the display unit of the user terminal 100. When a search query is input, information indicating the search query is stored in the user terminal 100 in a state associated with the user ID. The history information indicating the past search query (hereinafter also simply referred to as “search query history”) stored in association with the user ID in the user terminal 100 in this manner, for example, information indicating the search query is stored (searched). Each time the query history is updated), it is transmitted from the user terminal 100 to the integrated server 300. The search history acquisition unit 301 receives and acquires the search query history transmitted from the user terminal 100 to the integrated server 300.

  In the present embodiment, when a search query is stored in the user terminal 100, information indicating the date and time when the search query is input is stored in association with the search query history. Then, information indicating the date and time is transmitted from the user terminal 100 to the integrated server 300 together with the search query history. When acquiring the search query history as described above, the search history acquisition unit 301 receives and acquires information indicating the date and time together with the search query history. Note that the timing at which the search query history and the information indicating the date and time associated with the search query history are transmitted from the user terminal 100 to the integrated server 300 may not be the time at which the information indicating the search query is stored. May be transmitted.

  In addition, when one of the search result content list information acquired as a search result by the search query input on the content distribution service is clicked by the user, the content is included in the search result content. Information indicating the content (for example, metadata such as title and category) is stored in the user terminal 100 in a state in which the user ID is associated with the search query used to acquire the search result content. The Information indicating the past clicked content (hereinafter simply referred to as “search click history”) stored in the user terminal 100 in association with the user ID and the search query in this way is, for example, clicked. Each time information indicating content is stored (search click history is updated), the information is transmitted from the user terminal 100 to the integrated server 300. The search history acquisition unit 301 receives and acquires the search click history transmitted from the user terminal 100 to the integrated server 300.

  In this embodiment, when information indicating the clicked content is stored in the user terminal 100, information indicating the date and time when the content is clicked is stored in association with the search click history. Then, information indicating the date and time is transmitted from the user terminal 100 to the integrated server 300 together with the search click history. When acquiring the search click history as described above, the search history acquisition unit 301 receives and acquires information indicating the date and time together with the search click history. Note that the timing at which the search click history and the information indicating the date and time associated with the search click history are transmitted from the user terminal 100 to the integrated server 300 may not be the time at which the information indicating the clicked content is stored. It may be transmitted at the timing.

  As described above, the search history acquisition unit 301 is information that associates the past search query with the category of the content selected by the user among the contents corresponding to the past search query for each of the plurality of user IDs. A certain multiple user history (hereinafter, also simply referred to as “multiple user history”) is acquired. The search history acquisition unit 301 outputs the acquired multiple user history to the search history information storage unit 302. The search history acquisition unit 301 may output to the search history information storage unit 302 every time a plurality of user histories are acquired, or may output it every predetermined time.

  The search history information storage unit 302 stores the multiple user history output from the search history acquisition unit 301. FIG. 3 is an explanatory diagram showing the contents of information stored in the search history information storage unit 302. As shown in FIG. 3, the search history information storage unit 302 stores, as a plurality of user histories, information indicating a user ID, a search query, clicked content, and date and time in association with each other. As described above, the user ID is a number or abbreviation for identifying the user terminal 100. In FIG. 3, the user ID is indicated by “User1”, for example. The information indicating the search query is the above-described search query history. The information indicating the clicked content is the search click history (title and category metadata) described above. The information indicating the date and time is information indicating the date and time associated with the search query history or the search click history as described above. In FIG. 3, when any information is clicked from the list information indicating the content search result, information indicating the date and time when the information is clicked (that is, the date and time associated with the search click history) is stored. If no information is clicked, information indicating the date and time when the search request is made (that is, the date and time associated with the search query history) is stored.

  For example, in FIG. 3, the user terminal 100 with the user ID “User1” makes a search request with the search query “Idol A”, and among the search result content list information based on the search query, the title “Song K” and It is shown that the information indicating the content associated with the metadata of the category “Music, Idol” was clicked on September 15, 2014. In FIG. 3, proper nouns are indicated by character strings using alphabets. For example, the search query “idol A” indicates that the search query is a proper noun “A” of the idol.

  Further, for example, in the column of the user ID “User6”, only the search query “Idle A” and the date and time searched for on September 13, 2014 by this search query are stored, and the metadata (title and title) of the clicked content is stored. Category) is not stored. This indicates that the user terminal 100 with the user ID “User6” made a search request with the search query “Idle A”, but did not click on any of the search result content list information based on the search query. ing. That is, FIG. 3 shows that the user terminal 100 with the user ID “User6” does not have a search click history.

  The search history information storage unit 302 stores the information every time the search history acquisition unit 301 outputs information. Therefore, the database shown in FIG. 3 is updated each time information is output by the search history acquisition unit 301. Therefore, FIG. 3 illustrates data of the user IDs “User1” to “User6”, but this is only an example, and further data by the user IDs “User1” to “User6” or other user IDs Data may be included.

  The query category dictionary 303 is a database in which a predetermined search query and information that associates the degree of correspondence with the predetermined search query for each content category are stored. In this embodiment, the query category dictionary 303 is created by a query category dictionary creation unit 304 described later.

  FIG. 4 is an explanatory diagram showing the contents of information stored in the query category dictionary 303. FIG. 4A shows an example of the query category dictionary 303. In the query category dictionary 303 shown in FIG. 4A, a predetermined search query, each category corresponding to the predetermined search query, and the degree of correspondence of each category corresponding to the predetermined search query with respect to the search query. The evaluation value to be shown (hereinafter simply referred to as “query category evaluation value”) is stored.

  In the query category dictionary 303 of FIG. 4A, the query category evaluation values of the categories “idol”, “music”, “movie”, and “image” corresponding to the search query “idol A” are “1. 0, “0.50”, “0.25”, and “0.25”. That is, in the query category dictionary 303 of FIG. 4A, the search query “idol A” has the highest degree of correspondence with the category “idol”, followed by the category “music”, followed by the category “video” and the category. The order of “image” indicates that the degree of correspondence decreases.

  The query category dictionary creation unit 304 creates the above-described query category dictionary 303 based on, for example, a plurality of user histories stored by the search history information storage unit 302. The query category dictionary creation unit 304 may create the query category dictionary 303 based on all the data included in the multiple user history, or only data that goes back for a preset period among the data included in the multiple user history, for example. The query category dictionary 303 may be created based on the above. The query category dictionary creation unit 304 refers to a plurality of user histories in the search history information storage unit 302 and extracts a predetermined search query and each category corresponding to the search query. Then, the query category dictionary creation unit 304 calculates a query category evaluation value as follows.

  The query category dictionary creation unit 304 counts how many times each category is extracted. The number of times each category has been extracted indicates the number of times the content associated with each category has been clicked (hereinafter simply referred to as “number of clicks”). For example, the query category dictionary creation unit 304 refers to the plurality of user histories shown in FIG. 3 and searches the search query “idol A” and the categories “idol”, “music”, and “video” associated with the search query. , And “image” are extracted. Then, the number of clicks of each category (the number of times the content associated with each category has been clicked) is counted.

  FIG. 4B shows the number of clicks X of each category calculated in the process of creating the query category dictionary 303 and the maximum number of clicks Max (X) indicating the maximum number of clicks X. Yes. For example, in the multiple user history shown in FIG. 3, there are four categories “idle” corresponding to the search query “idle A”, so the category “idle” is extracted four times. Therefore, the click number X of the category “idle” is counted as “4”. Similarly, the click number X of the category “music” corresponding to the search query “idol A” is “2”, the click number X of the category “video” corresponding to the search query “idol A” is “1”, and the search query. The number of clicks X of the category “image” corresponding to “idle A” is counted as “1”. Therefore, the maximum click number Max (X) among the click number X of each category corresponding to the search query “idol A” is the click number X of the category “music”.

The query category dictionary creation unit 304 uses the number of clicks counted for each category as shown in FIG. 4B to calculate a query category evaluation value for each category based on the following equation (1). . The query category evaluation value is calculated for all categories corresponding to the search query.
Y (i) = X (i) / max (X (j)) (1)
However, 1 ≦ j ≦ n
Y (i): Query category evaluation value of i-th category X (i): Number of clicks of i-th category n: Number of category types

  For example, when the query category evaluation value of each category corresponding to the search query “idol A” is calculated, the query category dictionary 303 corresponds to the search query “idol A” as shown in FIG. The query category evaluation value of each category is calculated by dividing the number of clicks of each category by the number of clicks “4” of “music” having the maximum number of clicks in each category. As a result, the query category evaluation values of the categories “idol”, “music”, “movie”, and “image” corresponding to the search query “idol A” are as shown in FIG. “1.0”, “0.50”, “0.25”, and “0.25” are calculated, respectively.

  The query category dictionary creation unit 304 performs the above process for each search query included in the plurality of user histories in the search history information storage unit 302 (repeats the above process by changing the search query). That is, for all search queries included in a plurality of user histories, each category associated with the search query is extracted, and the query category evaluation value of each extracted category is calculated by the above-described calculation method. Note that the query category dictionary creation unit 304 does not have to calculate query category evaluation values for all search queries included in a plurality of user histories. For example, a query category evaluation value may be calculated only for a search query set in advance in the integrated server 300 by a user who is a user of the content distribution service, a business operator who is a provider of the service, or the like.

Note that the evaluation value of each category may be calculated by, for example, the following formula (2) instead of the above formula (1).
Y (i) = {X (i)} ² / {max (X (j))} ² (2)
According to the formula (2), when the number of clicks of the extracted category is small compared to the formula (1), the query category evaluation value of the category is calculated as a smaller value. For this reason, when using formula (2), there is an effect of moderating the upward trend of the query category evaluation value of the category with a small number of clicks. Therefore, the formula (2) is suitable when the category having a larger number of clicks is desired to have a higher degree of correspondence with the search query.

Moreover, you may calculate the query category evaluation value for every category instead of the said Numerical formula (1), for example by Numerical formula (3) shown below.
Y (i) = {X (i) / max (X (j))} ^1/2 (3)
According to the formula (3), in contrast to the formula (2), the query category evaluation value of the category is larger even when the number of clicks of the extracted category is smaller than the formula (1). Is calculated as This has the effect of moderating the upward trend of the query category evaluation value when the number of clicks is large. Therefore, Equation (3) is suitable when it is desired to provide a degree of correspondence with the search query even in a category with a small number of clicks.

  The query category dictionary creation unit 304 stores the query category evaluation value calculated as described above as a database in association with each search query and category. Thereby, the query category dictionary 303 is created.

  Note that the query category dictionary creation unit 304 may calculate the query category evaluation value of each category corresponding to the search query each time a plurality of user histories in the search history information storage unit 302 is updated. You may go to That is, the query category dictionary creation unit 304 may update the query category dictionary 303 every time the multiple user history is updated, or the query category dictionary 303 every predetermined period at a timing different from the update of the multiple user history. May be updated.

  The user feature vector calculation unit (calculation unit) 305 calculates a user feature amount indicating the degree of correspondence of the preference for each user ID with respect to the content category. The user feature amount indicating the degree of preference correspondence for each user ID with respect to the content category is information indicated as a vector amount including an evaluation value for each category indicating how much content the user is interested in to which category of content. It is. Hereinafter, the user feature quantity indicating the degree of preference for each user ID with respect to the content category is simply referred to as a “user feature vector”. The user feature vector calculation unit 305 outputs the calculated user feature vector to the user feature vector storage unit 306. The user feature vector storage unit 306 stores the output user feature vector.

  FIG. 5 is an explanatory diagram showing the contents of information stored in the user feature vector storage unit 306. FIG. 5 shows an example of a database in which the user feature vector evaluation value x ′ of the above-described user ID “UserA” is stored. The user feature vector evaluation value x ′ is an evaluation value for each category indicated by the user feature vector. In the database of FIG. 5, each category associated with the user ID “UserA” and a user feature vector evaluation value x ′ for each category are stored. Each category associated with the user ID “UserA” indicates a category to which the interest of the user ID “UserA” is directed. The user feature vector evaluation value x ′ for each category is an index for evaluating the degree of interest of each category to which the interest of the user ID “UserA” is directed.

  In the present embodiment, the larger the user feature vector evaluation value x ′, the higher the degree of interest of the user for each category. In the database of FIG. 5, each category “idol”, “music”, “video”, “image”, “comic”, “movie”, and “anime” to which the interest of the user ID “UserA” is directed. When the user feature vector evaluation value x ′ is compared, the user feature vector evaluation value x ′ of the category “comic” is the largest. Therefore, it is indicated that the degree of interest of the user ID “UserA” is the highest for the category “comic”.

  The user feature vector calculation unit 305 calculates a user feature vector as shown in FIG. 5 using, for example, a search query history. The user feature vector calculation unit 305 extracts a query category evaluation value (information associated with a correspondence degree corresponding to a past search query for each content category) from the query category dictionary 303, and uses the extracted query category evaluation value. Based on this, a user feature vector is calculated. More specific description will be given below.

  First, the user feature vector calculation unit 305 refers to a plurality of user histories in the search history information storage unit 302 and extracts a search query history of the user terminal 100 that is a calculation target of the user feature vector. The user feature vector calculation unit 305 extracts a query category evaluation value of each category corresponding to the past search query indicated by the extracted search query history from the query category dictionary 303. The user feature vector calculation unit 305 calculates a user feature vector based on the extracted query category evaluation value.

  In the present embodiment, the user feature vector calculation unit 305 refers to a plurality of user histories in the search history information storage unit 302 and extracts a search click history in a predetermined user terminal 100. The user feature vector calculation unit 305 may use this search click history for calculating the user feature vector.

  The user feature vector calculation unit 305 extracts, for example, a newly stored search query history and search click history from the previous extraction to the current extraction every predetermined period. Then, the user feature vector calculation unit 305 calculates a user feature vector using the search query history and the search click history extracted every predetermined period. Note that the user feature vector calculation unit 305 may calculate the user feature vector by extracting the search query history and the search click history each time the search query history and the search click history are updated.

The user feature vector calculation unit 305 calculates an evaluation value (hereinafter referred to as “user feature vector evaluation value”) for each category indicated by the user feature vector, for example, using the following mathematical formula (4). Note that the weighting factor α related to x, the weighting factor β related to y, and the weighting factor γ related to z are preset and stored in the integrated server 300. These coefficients can usually be set as various values depending on which of x, y, and z is to be emphasized. For example, these coefficients are set as positive values.
x ′ = αx + βy + γz (4)
However, when y = z = 0, x ′ = x
x ′: New user feature vector evaluation value x: Existing user feature vector evaluation value y: User feature vector evaluation value based on search query history z: User feature vector evaluation value based on search click history α: Weight coefficient related to x β: Weight coefficient related to y γ: Weight coefficient related to z

  That is, the user feature vector calculation unit 305 creates a new feature based on the existing user feature vector evaluation value x, the user feature vector evaluation value y based on the search query history, and the user feature vector evaluation value z based on the search click history. A user feature vector evaluation value x ′ is calculated. Note that the existing user feature vector evaluation value x is derived by the above formula (4) or set in advance before the calculation of the new user feature vector evaluation value x ′ this time, so that each user ID has already been set. Is a user feature vector evaluation value associated with.

  The user feature vector evaluation value y based on the search query history (hereinafter referred to as “query evaluation value y”) is the degree of correspondence of the preference for each user ID with respect to the content category, and the past search query and its category. This value is evaluated based on the degree of correspondence. The user feature vector evaluation value z (hereinafter referred to as “click evaluation value z”) based on the search click history is the degree of correspondence of the preference for each user ID with respect to the content category, and the content with which the category is previously associated. It is a value to be evaluated based on whether or not the indicated information is clicked.

  The query evaluation value y is calculated, for example, by extracting each query category evaluation value for each category associated with the search query from the query category dictionary 303 for a past search query for each user ID. The click evaluation value z is calculated by extracting a category associated with the clicked content from the search results based on the search click history for each user ID. A detailed method for calculating the query evaluation value y and the click evaluation value z will be described later with reference to FIGS. 6 and 7 and more specific examples.

  The query evaluation value y and the click evaluation value z are calculated without depending on each other, as indicated by the independent addition terms in the equation (4). Of the query evaluation value y and the click evaluation value z, it is sufficient that at least the query evaluation value y is calculated, and the click evaluation value z does not have to be calculated. That is, the new user feature vector evaluation value x ′ can be calculated without using the click evaluation value z. Therefore, for example, a new user feature vector evaluation value x ′ can be calculated even for a user who does not have a search click history, such as the user ID “User6” shown in FIG. 3.

  In addition, when calculating a new user feature vector evaluation value x ′, the user feature vector calculation unit 305 corresponds to one of the categories of the new user feature vector evaluation value x ′ calculated by Equation (4). When the evaluation value exceeds a predetermined threshold value (“1.0” in the present embodiment), normalization is performed. That is, normalization is performed by dividing the user feature vector evaluation value x ′ for each category calculated by the above equation (4) by the maximum value among the evaluation values exceeding the predetermined threshold, and the result Is a new user feature vector evaluation value x ′.

  Hereinafter, an example of a calculation process of the user feature vector evaluation value x ′ using the above equation (4) by the user feature vector calculation unit 305 will be described in detail with reference to FIGS. 6 and 7.

  As an example, the case where two search actions are performed in the user terminal 100 with the user ID “UserA” will be described. In the first search action, a search request is made by the search query “Idol A”, and the title “Song K”, the category “Music”, and “Idle” are included as metadata in the list information of the search result contents by the search query. Is clicked on information indicating a content that is previously associated. In the second search action, a search request is made by the search query “Comic S”, and the title “Comic S Comic Monochrome” and the category “Comic” are included as metadata in the list information of the search result contents by the search query. The information indicating the content associated with is clicked.

  Here, in the calculation of the following user feature vector evaluation value x ′, the database shown in FIG. 4 is used as the query category dictionary 303. In the above mathematical formula (4), α = 1.0, β = 0.5, and γ = 0.7. Further, it is assumed that the first search action is the first search action by “UserA”, and the user ID “UserA” does not have an existing user feature vector evaluation value x in advance.

  First, a method for calculating the user feature vector evaluation value x ′ when the first search action is performed will be described. FIG. 6 is a table showing the calculation process of the user feature vector evaluation value x ′ when the first search action is performed. In FIG. 6, each category associated with the search query “idol A”, an existing user feature vector evaluation value x for each category, a query evaluation value y that is a user feature vector addition unit based on a query history, and a click history A click evaluation value z, which is a user feature vector addition unit, and a new user feature vector evaluation value x ′ reflecting the first search action are stored.

  As described above, the first search action is the first search action by the user ID “UserA”, and the user ID “UserA” does not have the existing user feature vector evaluation value x. The user feature vector evaluation value x is an initial value “0” for any category. The query evaluation value y and the click evaluation value z shown in FIG. 6 are calculated by the user feature vector calculation unit 305 as follows.

  A method of calculating the query evaluation value y of the user ID “UserA” when the first search action is performed will be described. First, the user feature vector calculation unit 305 refers to the query category dictionary 303 regarding the past search query “idol A” of the user ID “UserA” and associates the category (“ Each query category evaluation value is extracted for each of “Idle”, “Music”, “Movie”, and “Image”). The user feature vector calculation unit 305 sets each extracted query category evaluation value as a query evaluation value y. The user feature vector calculation unit 305 sets the value “0” as the query evaluation value y for other categories not associated with the search query “idol A”. In this way, the query evaluation value y for each category of the user ID “UserA” reflecting the first search action is calculated.

  A method of calculating the click evaluation value z of the user ID “UserA” when the first search action is performed will be described. First, the user feature vector calculation unit 305 extracts a category associated with the content indicated by the information clicked by the user from the list information of the search result content based on the search click history of the user ID “UserA”. For example, since the category “music” and the category “idol” are associated in advance with the content clicked in the first search action, these categories are extracted. For each of the extracted categories, the user feature vector calculation unit 305 sets a predetermined constant (“1.0” in this embodiment) stored in advance in the integrated server 300 as the click evaluation value z for each category. To do. The user feature vector calculation unit 305 sets the value “0” as the click evaluation value z for each category for other categories other than the extracted category. In this way, the click evaluation value z for each category of the user ID “UserA” reflecting the first search action is calculated.

Based on the query evaluation value y and the click evaluation value z calculated in this way, the user feature vector calculation unit 305 calculates a new user feature vector evaluation value x ′ according to Equation (4). That is, the following calculation is performed for each category. However, “x ′ (category)” below indicates a new user feature vector evaluation value for the category. In the calculation process of this description, the fourth digit after the decimal point is rounded off, but the number of significant digits may be arbitrarily determined in the system according to the present embodiment.
x ′ (idle) = (1.0 × 1.0) + (0.5 × 1.0) + (0.7 × 1.0) = 1.20
x ′ (music) = (1.0 × 0.0) + (0.5 × 0.5) + (0.7 × 1.0) = 0.95
x ′ (video) = (1.0 × 0.0) + (0.5 × 0.25) + (0.7 × 0.0) = 0.125
x ′ (image) = (1.0 × 0.0) + (0.5 × 0.25) + (0.7 × 0.0) = 0.125

In the above case, the user feature vector evaluation value x ′ of the category “idle” exceeds the predetermined threshold “1.0”. Therefore, the user feature vector calculation unit 305 performs normalization by dividing the user feature vector evaluation value x ′ for each category calculated by the above equation (4) by the user feature vector evaluation value x ′. The result is set as a new user feature vector evaluation value x ′. Thereby, a new user feature vector evaluation value x ′ is calculated as follows.
x '(idol) = 1.0
x '(music) = 0.792
x '(video) = 0.104
x ′ (image) = 0.104

  Next, a user feature vector calculation method when the second search action is performed will be described. FIG. 7 is a table showing a calculation process of the user feature vector evaluation value x ′ when the second search action is performed. In FIG. 7, each category associated with the search query “comic S”, an existing user feature vector evaluation value x, a query evaluation value y, a click evaluation value z, and a new user feature vector evaluation value for each category. x ′ is stored.

  Since this is the second search action, the user ID “UserA” has the existing user feature vector evaluation value x obtained by the first search action. That is, the existing user feature vector evaluation value x shown in FIG. 7 is the user feature vector evaluation value x ′ shown in FIG. 6. Similar to the first search action described above, the query evaluation value y and the click evaluation value z shown in FIG. 6 are calculated by the user feature vector calculation unit 305 as follows.

  A method of calculating the query evaluation value y of the user ID “UserA” when the second search action is performed will be described. First, the user feature vector calculation unit 305 refers to the query category dictionary 303 regarding the past search query “comic S” of the user ID “UserA”, and associates the category (“ Each query category evaluation value is extracted for each of “comic”, “moving image”, “movie”, and “animation”. As in the case where the first search action is performed, the user feature vector calculation unit 305 sets each extracted query category evaluation value as the query evaluation value y of these categories, and the query evaluation values of other categories. The value “0” is set as y. In this way, the query evaluation value y for each category of the user ID “UserA” reflecting the second search action is calculated.

  A method of calculating the click evaluation value z of the user ID “UserA” when the second search action is performed will be described. First, the user feature vector calculation unit 305 extracts a category associated with the content indicated by the information clicked by the user from the list information of the search result content based on the search click history of the user ID “UserA”. For example, since the category “comic” is associated in advance with the content clicked in the second search action, this category is extracted. As in the case where the first search action is performed, the user feature vector calculation unit 305 sets a predetermined constant “1.0” as the click evaluation value z of the extracted category, and clicks on other categories A value “0” is set as the evaluation value z. In this way, the click evaluation value z for each category of the user ID “UserA” reflecting the second search action is calculated.

As in the case of the first search action, the user feature vector calculation unit 305 calculates a new user feature vector evaluation value x ′ for each category according to the equation (4). That is, the following calculation is performed for each category.
x ′ (idle) = (1.0 × 1.0) + (0.5 × 0.0) + (0.7 × 0.0) = 1.0
x '(music) = (1.0 x 0.792) + (0.5 x 0.0) + (0.7 x 0.0) = 0.792
x ′ (video) = (1.0 × 0.104) + (0.5 × 0.75) + (0.7 × 0.0) = 0.479
x ′ (image) = (1.0 × 0.104) + (0.5 × 0.0) + (0.7 × 0.0) = 0.104
x '(comic) = (1.0 x 0.0) + (0.5 x 1.0) + (0.7 x 1.0) = 1.2
x ′ (movie) = (1.0 × 0.0) + (0.5 × 0.5) + (0.7 × 0.0) = 0.25
x '(animation) = (1.0 x 0.0) + (0.5 x 0.25) + (0.7 x 0.0) = 0.125

In the above case, the user feature vector evaluation value x ′ (comic) of the categories “comic” and “idol” exceeds a predetermined threshold “1.0”, and among them, the user feature vector evaluation of the category “comic” The value x ′ is the maximum value. Therefore, the user feature vector calculation unit 305 divides the user feature vector evaluation value x ′ for each category calculated by the above equation (4) by the value of the user feature vector evaluation value x ′ of the category “comic”. Then, normalization is performed, and the result is set as a new user feature vector evaluation value x ′. Thereby, a new user feature vector evaluation value x ′ is calculated as follows.
x '(idol) = 0.833
x '(music) = 0.66
x '(video) = 0.399
x '(image) = 0.087
x '(comic) = 1.0
x '(movie) = 0.208
x '(animation) = 0.104

  In the above example, the user feature vector calculation unit 305, when there is one history information extracted every predetermined time (search query and click newly stored from the previous extraction to the current extraction) However, the present invention is not limited to this. That is, there may be a plurality of pieces of history information extracted every predetermined time (a plurality of search queries and clicked contents newly stored from the previous extraction to the current extraction). In this case, the user feature vector evaluation value x ′ is calculated based on a plurality of history information. Note that details of the calculation method of the query evaluation value y and the click evaluation value z when the user feature vector evaluation value x ′ is calculated based on a plurality of history information will be described later with reference to the flowcharts of FIGS. 13 and 14. . The user feature vector calculation unit 305 outputs the user feature vector evaluation value x ′ for each user ID calculated as described above to the user feature vector storage unit 306.

In the above equation (4), s = y + z is defined as the sum of the query evaluation value y and the click evaluation value z, and a new user feature vector evaluation value x ′ is calculated by the following equation (5). Also good.
x '= x (1 + s ) ···· (5) ( However, in the case of x = 0, the _{x 0} as a predetermined _{constant, x' = x 0 (1} + s)
In the method using Equation (5), the sum of the query evaluation value y and the click evaluation value z is multiplied by the existing user feature vector evaluation value x, so that the new user feature vector evaluation value x ′ is likely to fluctuate. This is suitable when the latest search query history and search click history are to be preferentially affected. The case where x = 0 is a case where there is no user feature vector related to the category. As the predetermined constant x ₀ in this case, for example, a value calculated based on a user feature vector related to another category may be used. Specifically, it may be a predetermined constant x ₀ the mean or median value or the like of each user feature vector evaluation value x calculated for other categories. Also, if another user characteristic vector evaluation value x calculated for the category is not present at all, that is, when there is no existing user characteristic vector may be 0.5 as an initial value as a predetermined constant x _0.

  The search request receiving unit 307 acquires a user ID and a search query from the user terminal 100 and receives a content search request. Specifically, when the user ID is input by the user on the login screen displayed on the display unit of the user terminal 100 as a search query, information indicating the user ID is transmitted from the user terminal 100 to the integration server 300. The The search request receiving unit 307 receives and acquires the user ID transmitted from the user terminal 100 to the integrated server 300. Further, when the user inputs the search window on the content distribution service, information indicating the search query is transmitted from the user terminal 100 to the integrated server 300. The search request receiving unit 307 receives and acquires information indicating a search query transmitted from the user terminal 100 to the integrated server 300.

  The search request reception unit 307 outputs information indicating the acquired search query to the search request transmission unit 308. The search request transmission unit 308 transmits information indicating the search query output from the search request reception unit 307 to the search server 400.

  The search result content acquisition unit 309 receives and acquires the search result content corresponding to the search query from the search request transmission unit 308 from the search server 400. Content metadata (title and category) is associated in advance with the content included in the search result content (see FIG. 8A). The search result content acquisition unit 309 outputs the acquired search result content to the search result content output unit 311.

  The rank calculation unit 310 determines an order for rearranging the list information of the search result content based on the user feature vector calculated by the user feature vector calculation unit 305. The rank calculation unit 310 refers to the database stored in the user feature vector storage unit 306, extracts the user feature vector evaluation value x ′ for each category for the calculation target user, and based on the user feature vector evaluation value x ′. The order for rearranging the list information of the search result contents is determined. Hereinafter, the ranking of the contents arranged in this order is also simply referred to as “content ranking”. For example, the rank calculation unit 310 sets the category order for the corresponding user ID associated with the user feature vector so that the category with the larger user feature vector evaluation value x ′ is higher (hereinafter referred to as “category priority”). (Refer to S71 of FIG. 8B and FIG. 15). The rank calculation unit 310 determines the content rank based on the priority of the category (see (c) and (d) of FIG. 8 and S72 of FIG. 15). The rank calculation unit 310 outputs information indicating the determined content rank to the search result content output unit 311.

  The search result content output unit 311 outputs the search result content list information output from the search result content acquisition unit 309 to the user terminal 100 based on the user feature vector calculated by the user feature vector calculation unit 305. The search result content output unit 311 rearranges the search result content list information output from the search result content acquisition unit 309 based on, for example, information indicating the content order output by the rank calculation unit 310. That is, the search result content output unit 311 rearranges the search result content list information and outputs it to the user terminal 100 so that each content included in the search result content follows the content order. Note that the category priority and the content ranking may be calculated by the search result content output unit 311 instead of the rank calculation unit 310.

  Hereinafter, a method of rearranging the search result content list information from the search result content acquisition unit 309 based on the user feature vector by the rank calculation unit 310 will be described in detail with reference to FIG. FIG. 8 is an explanatory diagram illustrating a method of rearranging search result content list information from the search result content acquisition unit 309 based on user feature vectors. The example of FIG. 8 shows a case where the user with the user ID “UserA” makes a search request with the search query “Comic R”.

  FIG. 8A is a table showing item numbers and metadata (title and category) associated with each content included in the search result content acquired by the search result content acquisition unit 309. The item number associated with the search result content is an identification number of the content included in the search result content. The item number corresponds to, for example, a priority order indicating which information indicating which content is arranged at the top when being arranged as list information on the display unit of the user terminal 100.

  FIG. 8B is a table showing the category priorities calculated by the rank calculation unit 310 based on the user feature vectors calculated by the user feature vector calculation unit 305 (see S71 in FIG. 15). As shown in FIG. 8B, the priority of each category indicated by the feature vector in the user ID “UserA” is “image”, “animation”, “movie”, “movie”, “music”, “ It is calculated in the order of “Idol” and “Comic”.

  As shown in FIG. 8C, the search result content output unit 311 associates the category associated with the content in FIG. 8A with the category in FIG. 8B. Associate the item number of the content with the priority of the category. For example, in (a) of FIG. 8, the category “comics” is associated with the contents of the item numbers “5” and “6”. Therefore, by associating the category “comic” with the category “comic” in FIG. 8B, the item numbers “5” and “6” of the contents are associated with the priority of the category “comic”. Yes. In addition, the category “movie” and “moving image” are associated with the content of the item number “1” in FIG. As described above, when there are a plurality of categories associated with the content, for example, a category having a higher priority is associated. In other words, by matching the category “moving image” having a higher priority among “movie” and “moving image”, the item number “1” of the content is associated with the priority of the category “moving image”.

  FIG. 8D is a table showing the result of rearranging the item numbers of contents in accordance with the priorities of the categories associated as described above. As shown in FIG. 8D, the item numbers “1” to “6” included in the search result content are rearranged according to the priority of the category associated with each item number ( (See S72 in FIG. 15). As a result, the content order of the content of item number “1” is rearranged according to the priority of the category, and is the fifth from the top. The item numbers “5” and “6” are in the same order according to the priority of the associated category. In this case, they are rearranged according to the order of the original item numbers. Accordingly, the content order of the content of item number “5” is the first from the top, and the content order of content of the item number “6” is the second from the top.

  Next, the functional configuration of the search server 400 (search means) will be described with reference to FIG. 1 again. As shown in FIG. 1, the search server 400 includes a query acquisition unit 401, a content calculation unit 402, and a search result content transmission unit 403.

  The query acquisition unit 401 receives and acquires information indicating the search query transmitted from the search request transmission unit 308 of the integrated server 300. The query acquisition unit 401 outputs information indicating the acquired search query to the content calculation unit 402.

  Based on the information indicating the search query from the query acquisition unit 401, the content calculation unit 402 calculates the search result content that hits the search request by the search query from the content stored in the search target database. The content calculation unit 402 outputs the calculated search result content to the search result content transmission unit 403.

  The search result content transmission unit 403 transmits the search result content output by the content calculation unit 402 to the search result content acquisition unit 309 in the integrated server 300.

  Next, the operation of the content search result providing system 200 according to the present embodiment will be described with reference to FIGS. FIG. 9 is a sequence diagram showing an overall operation procedure of the content search result providing system 200 shown in FIG.

  As shown in FIG. 9, in the content search result providing system 200, a multi-user history is transmitted from the user terminal 100 to the integrated server 300 (S 1). Subsequently, in the integrated server 300, a query category dictionary 303 is created based on the multiple user history (S2). Subsequently, in the integrated server 300, a user feature vector is calculated (S3).

  Subsequently, a search query is transmitted from the user terminal 100 to the integrated server 300 (S4), and a search query is transmitted from the integrated server 300 to the search server 400 (S5). Subsequently, the search server 400 calculates search result content corresponding to the received search query (hits a search request based on the search query) (S6). Subsequently, the search result content is transmitted from the search server 400 to the integrated server 300 (S7). Subsequently, in the integrated server 300, the order for rearranging the list information of the search result contents (that is, the order of the contents included in the search result contents) is calculated (S8). Subsequently, the search result contents rearranged (sorted) from the integrated server 300 to the user terminal 100 are output as list information (S9).

  Hereinafter, a more detailed operation procedure will be described. FIG. 10 is a flowchart showing an overall operation procedure of the content search result providing system 200 shown in FIG.

  As illustrated in FIG. 10, in the content search result providing system 200, the search history acquisition unit 301 acquires the history (search query and information indicating the clicked content) transmitted from the user terminal 100, and the acquired search. The query history and search click history are output to the search history information storage unit 302 (S10: search history acquisition step).

  Subsequently, the query category dictionary creation unit 304 creates a query category dictionary 303 based on the plurality of user histories in the search history information storage unit 302 (S20). Details of S20 will be described later.

  Subsequently, the user feature vector calculation unit 305 extracts a query category evaluation value from the query category dictionary 303 using a past search query, and calculates a user feature vector based on the extracted query category evaluation value ( S30: User feature amount calculation step). Details of S30 will be described later.

  Subsequently, the search request accepting unit 307 receives and acquires the user ID and the search query from the user terminal 100, and accepts a content search request (S40). Subsequently, the search request transmission unit 308 transmits information indicating the search query to the query acquisition unit 401, and the content calculation unit 402 calculates the search result content (S50).

  Subsequently, the search result content transmission unit 403 transmits the search result content, and the search result content acquisition unit 309 receives and acquires the search result content (S60: search result acquisition step). Subsequently, the content ranking of the content included in the search result content is calculated by the ranking calculation unit 310 using the user feature vector for each user ID (S70). Details of the process of S70 will be described later. Subsequently, the search result content output unit 311 outputs the search result content list information arranged in accordance with the content order to the user terminal 100 (S80: output step).

  Next, details of a method for creating the query category dictionary 303 by the query category dictionary creation unit 304 will be described with reference to FIG. FIG. 11 is a flowchart showing a detailed processing method of S20 shown in FIG.

  As illustrated in FIG. 11, first, a plurality of user histories stored in the search history information storage unit 302 are associated with the clicked content among the content included in the search result content based on the predetermined search query and the search query. A category is extracted, and the number of clicks for each category is calculated (S21). Subsequently, using the number of clicks for each category, the query category evaluation value of the i-th category is calculated by the above formula (1) (S22). The i-th “i” indicates an index (subscript) of a category associated with the clicked content among the contents included in the search result content by the predetermined search query. When the number of types of categories associated with the clicked content is n, the process of S22 is repeated from i = 1 to i = n. That is, the process of S22 is performed for all categories associated with the content. The processes of S21 to S22 are performed for all search queries set in advance.

  Next, details of a user feature vector calculation method by the user feature vector calculation unit 305 will be described with reference to FIG. FIG. 12 is a flowchart showing a detailed processing method of S30 shown in FIG.

  As shown in FIG. 12, first, a user ID that is a target for calculating a user feature vector is selected (S31). Subsequently, for the selected user ID, a query evaluation value y is calculated (S32), and a click evaluation value z is calculated (S33). Details of the calculation method of the query evaluation value y and the click evaluation value z will be described later with reference to FIGS. 13 and 14.

  Subsequently, based on the query evaluation value y calculated in S32 and the click evaluation value z calculated in Step S33, a user feature vector evaluation value x 'is calculated by Step Equation (4) (S34). Subsequently, it is determined whether or not the user feature vector evaluation value x ′ of any category is larger than 1.0 (S35). When it is determined that the user feature vector evaluation value x ′ is 1.0 or less (S35; NO), the process proceeds to S37 without proceeding to S36. That is, in this case, the user feature vector evaluation value x ′ calculated in S34 is the user feature vector evaluation value x ′ relating to the user ID selected in S31 without performing the normalization process in S36 described later. Then, the process proceeds to S37.

  When it is determined that the user feature vector evaluation value x ′ is greater than 1.0 (S35; YES), the process proceeds to S36. In the process of S36, the user feature vector evaluation value x 'is normalized. That is, normalization is performed by dividing by the maximum value of the evaluation value x ′, and the result is set as the user feature vector evaluation value x ′. Then, the process proceeds to S37. In the process of S37, it is determined whether or not the next user ID exists in the multiple user history. The next user ID indicates a user ID for which a user feature vector has not been calculated. If the next user ID exists (S37; YES), the next user ID is selected in S31, and then the processes of S32 to S37 are performed. If the next user ID does not exist (S37; NO), the process ends.

  Next, details of a method for calculating the query evaluation value y and the click evaluation value z will be described with reference to FIGS. In the following calculation method, when there are a plurality of history information extracted by the user feature vector calculation unit 305 every predetermined time, the query evaluation value y and the click evaluation value z are calculated based on the plurality of history information. explain. That is, the query evaluation value y and the click when there are a plurality of history information extracted every predetermined time (when there are a plurality of search queries and clicked contents newly stored from the previous extraction to the current extraction) A method for calculating the evaluation value z will be described. FIG. 13 is a flowchart showing a detailed processing method of S32 shown in FIG. That is, it is a flowchart for explaining a detailed calculation method of the query evaluation value y.

  As shown in FIG. 13, first, from the history information extracted by the user feature vector calculation unit 305 for the user ID that is the target of calculation of the user feature vector selected in S31 in FIG. Selected (S41). Here, the next search query indicates a search query that is not used in the calculation process of the user feature vector among the history information extracted for the same user ID as the calculation target. Subsequently, it is determined whether or not the selected search query exists in the query category dictionary 303 (S42). If the selected search query does not exist in the query category dictionary 303 (S42; NO), the process returns to S41 and the next search query is selected from the history information extracted again. When the search query exists in the query category dictionary 303 (S42; YES), the process proceeds to S43.

  In the process of S43, the category corresponding to the search query selected in S41 and the query category evaluation value of each category are extracted from the query category dictionary 303, and the extracted query category evaluation value is set to p1 (S43). ). Subsequently, p1 in S43 is set as an evaluation value of each category extracted from the query category dictionary 303 (S44). In S44, when the category extracted from the query category dictionary 303 already exists and the evaluation value of the category is already set, p1 in S43 is added to the evaluation value of the category, and a new value is added. The evaluation value of the category.

  Subsequently, for the same user ID, it is determined whether or not the next search query exists in the extracted history information (S45). When the next search query exists (S45; YES), the process returns to S41 again, and the next search query is selected. Subsequently, the processes of S42 to S45 are performed. When the next search query does not exist (S45; NO), that is, when the processes of S41 to S44 are performed for all of the plurality of search queries included in the extracted history information, the process proceeds to S46. In the process of S46, the value determined as the category evaluation value in S44 is set as the query evaluation value y of each category, and the calculation process of the query evaluation value y ends.

  FIG. 14 is a flowchart showing a detailed processing method of S33 shown in FIG. That is, it is a flowchart for explaining a detailed calculation method of the click evaluation value z.

  As shown in FIG. 14, the next search click history is first extracted from the history information extracted by the user feature vector calculation unit 305 for the user ID that is the calculation target of the user feature vector selected in S <b> 31 in FIG. 12. Selected (S51). Here, the next search click history indicates a search click history that is not used in the calculation process of the user feature vector among the history information extracted for the same user ID. Subsequently, the constant p2 (1 in the present embodiment) is set as the evaluation value of each category associated with the clicked content indicated by the selected search click history (S52). If a category associated with the selected content already exists and an evaluation value of the category has already been set, a constant p2 is added to the evaluation value of the category, and the evaluation value of the new category Is done.

  Subsequently, it is determined whether or not the next search click history exists in the extracted history information with the same user ID (S53). When the next search click history exists (S53; YES), the process returns to S51 again, and the next search click history is selected. Subsequently, the processing of S52 to S53 is performed. When the next search click history does not exist (S53; NO), that is, when the processes of S51 to S53 are performed for all of the plurality of search click histories included in the extracted history information, the process proceeds to S54. In the processing of S54, the value determined as the category evaluation value in S54 is set as the click evaluation value z of each category, and the calculation processing of the click evaluation value z ends.

  Next, with reference to FIG. 15, the details of a method for calculating the content rank of the content included in the search result content using the user feature vector for each user ID by the rank calculation unit 310 will be described. FIG. 15 is a flowchart showing a detailed processing method of S70 shown in FIG.

  As shown in FIG. 15, first, the user feature vectors calculated by the user feature vector calculation unit 305 are rearranged (sorted) in descending order from the largest value to the smallest value for the user feature vector evaluation value x ′. The category order of the rearranged user feature vectors is set as the category priority for the corresponding user ID (user feature vector calculation target user ID) (S71). Subsequently, in the search result content acquired by the search result content acquisition unit 309, the display order (content order of content included in the search result content) displayed on the display unit of the user terminal 100 based on the category priority obtained in S71. It is determined (S72). As described above, the content rank calculation processing by the rank calculation unit 310 ends.

  As described above, according to the content search result providing system 200 and the content search result providing method according to the present embodiment, the query category evaluation value associated with the degree of correspondence to the past search query for each content category is extracted from the query category dictionary 303. Then, based on the extracted query category evaluation value, a user feature vector indicating the degree of correspondence of the preference for each user with respect to the content category is calculated. Therefore, the user feature vector is calculated without using the search click history of the user who makes the search request. Since the search result of the content corresponding to the search query at the time of the search request is output based on the user feature vector, even if there is no search click history of the user who makes the search request, the user's preference is reflected. Search results can be output. As described above, even when there is no search click history of a user who makes a search request, it is possible to provide content that reflects the user's preference.

  Further, according to the content search result providing system 200 according to the present embodiment, the search result of the content in the order based on the user feature vector is output, so even when there is no search click history of the user who makes the search request, Search results arranged in order according to the user's preference can be output.

  Further, according to the content search result providing system 200 according to the present embodiment, the search server 400 can search for content corresponding to the search query at the time of the search request transmitted from the user terminal 100.

  Further, according to the content search result providing system 200 according to the present invention, the query category dictionary 303 is created based on a plurality of user histories in which a past search query and a category are associated for each of a plurality of users. The reliability of the information stored in 303 can be further increased.

  As mentioned above, although embodiment of this invention has been described, this invention is not necessarily limited to embodiment mentioned above, A various change is possible in the range which does not deviate from the summary.

  For example, in the above embodiment, the query category dictionary 303 is created by the query category dictionary creation unit 304, but the present invention is not limited to this. For example, the integrated server 300 may have a query category dictionary that is preset and stored.

DESCRIPTION OF SYMBOLS 200 ... Content search result provision system, 301 ... Search history acquisition part (search history acquisition means), 303 ... Query category dictionary, 304 ... Query category dictionary creation part, 305 ... User feature vector calculation part (user feature amount calculation means), 309 ... Search result content acquisition unit (search result acquisition unit), 310 ... Rank calculation unit (rank calculation unit), 311 ... Search result content output unit (output unit), 400 ... Search server (search unit), S10 ... Search history Acquisition step, S30 ... User feature amount calculation step, S60 ... Search result acquisition step, S80 ... Output step.

Claims (4)

A query category dictionary that stores information associating a predetermined search query with a degree of correspondence to the predetermined search query for each category of content;
Search history acquisition means for acquiring a search query history indicating a past search query transmitted from the user terminal;
Using the search query history, information that associates the degree of correspondence with the past search query for each category of the content is extracted from the query category dictionary, and the user for the content category is extracted based on the extracted information. User feature amount calculating means for calculating a user feature amount indicating a degree of preference for each preference;
Search result acquisition means for acquiring a search result of content corresponding to the search query at the time of the search request transmitted from the user terminal and including information indicating the content associated with the category;
Output means for outputting a search result of the content based on the user feature amount;
Equipped with a,
The predetermined search query includes the past search query,
The search history acquisition unit is configured to obtain, for each of a plurality of users, a plurality of user histories that are information in which the past search query is associated with a category of content selected by the user among contents corresponding to the past search query. Have acquired
A content search result providing system further comprising a creation unit that creates the query category dictionary based on the plurality of user histories . A rank calculating means for determining an order of information indicating the content included in the search result of the content based on the user feature amount;
The content search result providing system according to claim 1, wherein the output unit outputs the search result of the content according to the order determined by the rank calculation unit. The content search result providing system according to claim 1 or 2 , further comprising search means for searching for content corresponding to a search query at the time of a search request transmitted from the user terminal. A content search result providing method which is an operation method of a content search result providing system including a query category dictionary in which information in which a predetermined search query is associated with a degree of correspondence to the predetermined search query for each content category is stored. There,
A search history acquisition step for acquiring a search query history indicating a past search query transmitted from the user terminal;
Using the search query history, information that associates the degree of correspondence with the past search query for each category of the content is extracted from the query category dictionary, and the user for the content category is extracted based on the extracted information. A user feature amount calculating step for calculating a user feature amount indicating a degree of preference for each preference;
A search result acquisition step for acquiring a search result of content corresponding to the search query at the time of the search request transmitted from the user terminal and including information indicating the content associated with the category;
An output step of outputting a search result of the content based on the user feature amount;
Only including,
The predetermined search query includes the past search query,
In the search history acquisition step, for each of a plurality of users, a plurality of user histories that are information that associates the past search query with a category of the content selected by the user among the contents corresponding to the past search query. Have acquired
A content search result providing method , further comprising a creation step of creating the query category dictionary based on the multi-user history .

Images