JP2018072983A - Retrieval device, retrieval method and retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device, a method and a program that retrieve a content as intended by a user.SOLUTION: A retrieval device 100 has an extraction section 132, a calculation section 134, and a retrieval section 138. The extraction section 132 extracts features to be used for retrieval processing from a content. The calculation section 134 calculates a prescribed index value set for each feature on the basis of a cooccurrence relation on a combination of the features extracted by the extraction section 132, and associates the calculated index value with the content, then stores it in a prescribed storage section 120. The retrieval section 138 retrieves a content suitable for a query sent from a user on the basis of the calculated prescribed index value calculated by the calculation section 134.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a search device, a search method, and a search program.

  In recent years, with the development of communication networks, various services are provided via networks. In relation to this, various search techniques for searching for a service existing on a network or searching for a desired product or article in the service have been proposed.

  For example, a technique is known that can efficiently perform a process of searching (extracting) an advertisement to be distributed to a user by obtaining a predicted value of an advertisement index used for search at high speed.

JP2015-225549A

  However, with the above-described conventional technology, it is not always possible to efficiently search for content according to the user's intention. Specifically, the above-described conventional technology improves search efficiency by obtaining a predicted value of a predetermined index at a high speed, and is a process for searching for content according to the intention of a user who transmits a query. Not always applicable.

  The present application has been made in view of the above, and an object of the present invention is to provide a search device, a search method, and a search program that can efficiently search for content that matches a user's intention.

  The search device according to the present application is a predetermined set for each feature based on a co-occurrence relationship in an extraction unit that extracts features to be used for search processing from content and a combination of features extracted by the extraction unit. Based on the predetermined index value calculated by the calculation unit, the calculation unit that calculates the index value of the image, and stores the calculated index value and the content in association with each other in the predetermined storage unit And a search unit for searching for content that matches the query.

  According to one aspect of the embodiment, there is an effect that it is possible to efficiently search for content according to the user's intention.

FIG. 1 is a diagram (1) illustrating an example of a search process according to the embodiment. FIG. 2 is a diagram (2) illustrating an example of a search process according to the embodiment. FIG. 3 is a diagram illustrating a configuration example of the search system according to the embodiment. FIG. 4 is a diagram illustrating a configuration example of the search device according to the embodiment. FIG. 5 is a diagram illustrating an example of a content storage unit according to the embodiment. FIG. 6 is a diagram illustrating an example of a search information storage unit according to the embodiment. FIG. 7 is a diagram illustrating an example of an index value storage unit according to the embodiment. FIG. 8 is a diagram illustrating a configuration example of the web server according to the embodiment. FIG. 9 is a flowchart (1) illustrating a processing procedure according to the embodiment. FIG. 10 is a flowchart (2) illustrating a processing procedure according to the embodiment. FIG. 11 is a hardware configuration diagram illustrating an example of a computer that realizes the function of the search device.

  Hereinafter, embodiments for carrying out a search device, a search method, and a search program according to the present application (hereinafter referred to as “embodiments”) will be described in detail with reference to the drawings. Note that the search device, the search method, and the search program according to the present application are not limited by this embodiment. In addition, the embodiments can be appropriately combined within a range that does not contradict processing contents. In the following embodiments, the same portions are denoted by the same reference numerals, and redundant description is omitted.

[1. Example of search process)
First, an example of search processing according to the embodiment will be described with reference to FIGS. 1 and 2. FIG. 1 is a diagram (1) illustrating an example of a search process according to the embodiment. FIG. 1 shows an example of processing in which a predetermined index value used for search processing is calculated for content to be searched by the search device 100 according to the present application. In the embodiment, the content to be searched is an item on the auction site (more precisely, the content means data related to an introduction page for the exhibitor to introduce the item. In the following, the auction site. In the case of showing an exhibit itself such as a product in the above, it is expressed as “product related to content”).

  A search device 100 shown in FIG. 1 is a server device that holds content submitted by exhibitors. For example, the search device 100 works in conjunction with the web server 30 (see FIG. 3) that provides an auction site to search for items that have been exhibited on the auction site. That is, the search device 100 according to the embodiment receives a query from a user who uses an auction site, and transmits content that matches the query to the user as a search result for the received query. Specifically, when the search device 100 receives a content search request from the user terminal 10 (see FIG. 2), which is an information processing terminal operated by the user, the search device 100 matches the query from the held content. Search for content. Then, the search device 100 transmits the searched content to the user terminal 10 as a search result.

  A submission terminal 20 shown in FIG. 1 is an information processing terminal used by a seller who sells a product on an auction site. For example, the submission terminal 20 submits content to the search device 100 in accordance with an operation by the exhibitor. In the embodiment, the content submitted by the exhibitor is a data group including an introduction sentence (text data) for introducing a product (exhibit) associated with the content, image data obtained by capturing the product, and the like. . The content also includes the title of the content and information for identifying the exhibitor who has exhibited the content. In the embodiment, for the sake of simplification of description, an example in which the search device 100 performs a search using text data included in the content title is described. A search may be performed using text data or the like given as an image description.

  Here, when searching for content on an auction site or the like, it is desirable to return content that conforms to the query used for the search and that matches the user's intention to the user as a higher rank of the search results. For example, queries in search processing include queries that have different intentions even if they have the same notation. In such a case, it is desirable that the user's intention included in the query can be taken into account and reflected in the search results.

  For example, it is assumed that a user who wants to search for items on an auction site transmits a query combining two words “crab” and “shipping included” as a search query. In this case, usually, a method of calculating a correlation between a word included in the content to be searched and a query as an index value is employed. Then, the content is ranked based on the calculated score, and the search results are displayed on the user terminal 10 in the ranked order.

  However, there is a possibility that the items to be exhibited on the auction site include, for example, “crabs” as foods and “brake calipers” which are motorcycle articles similar to crab shapes. In this case, depending on the search method, content unintended by the user may be searched as a higher rank of the search result. For example, the user who performed the search sends a query combining two words “crab” and “shipping included” in an attempt to search for “crab” as food, but the “brake caliper” is a motorcycle article. "May be received as a search result.

  Conventionally, a method for returning a search result closer to the user's intention by using tf-idf (Term Frequency, Inverse Document Frequency) or the like has been proposed. However, when these are used as search methods, the expressiveness is weak to express the intention of the search indicated in the search query, and the search results ranked by these methods are accurate to reflect the user's intention. May not appear. Further, when the conventional method is used, since the word matching rate and the like are calculated after the query is transmitted from the user, it may be difficult to quickly return the search result to the user.

  Also, a method of performing a search using the co-occurrence degree (for example, cosine similarity) between a word included in the query and a word included in the content is also assumed. In this case, the more easily the words included in the query and the words included in the content co-occur, the higher the score indicating the search result, so that it is easier to derive a search result closer to the user's intention. However, also in this method, the co-occurrence degree of the word included in the query received from the user and the word included in the content must be calculated every time the query is received from the user. For this reason, in an auction site or the like where the amount of content to be searched for is large, the time spent for calculation becomes long, so it is practically difficult to adopt such a method.

  Therefore, the search device 100 according to the embodiment employs a method of calculating the co-occurrence degree of words included in the content in advance when the content to be searched is submitted and stored in the storage unit. With this technique, the search device 100 can accurately search for content that matches the intention of the user who transmitted the query, and can efficiently perform the search process quickly. Hereinafter, an example of a search process performed by the search device 100 will be described along the flow with reference to FIGS. 1 and 2.

  First, the exhibitor submits content to the search device 100 via the submission terminal 20 (step S01). For example, it is assumed that the exhibitor has submitted the content C01 and the content C02 shown in FIG. The content C01 is a content that uses crab as food as a product, and the title of the content is “I will deliver fresh crab from Hokkaido (shipping included)!”. The content C02 is a content that uses a brake caliper as a motorcycle article as a product, and the title of the content is “Brake caliper is sold (crab type, made of titanium, including shipping fee)”.

  When receiving a content submission from an exhibitor, the search device 100 extracts features to be used for search processing from the content at the stage of storing the content in a storage unit such as a memory. Specifically, as a feature, the search device 100 extracts a word that can be a query from words included in the content (step S02). Note that the search device 100 according to the embodiment uses “content title” for search processing, and therefore extracts a word that can be a query from text data included in the content title.

  A word that can be a query is appropriately set by an administrator of the search device 100 based on, for example, a result of learning the number of times it is actually used as a query. For example, the search device 100 performs a morphological analysis on the title of the content, and extracts a predetermined number (for example, four) of words that are frequently used as a query from the included words. Note that the search device 100 may extract a word that can be a query using a weight value such as tf-idf in the content to be searched, for example. For example, the search device 100 calculates a weight value such as tf-idf of each word with respect to a word subjected to morphological analysis in a predetermined content, and extracts words in order from the top of the calculated value. Accordingly, the search device 100 can easily extract, for example, a word unique to the content, and can exclude a word that is widely used from the extracted word.

  In the example of FIG. 1, for the content C01, the search apparatus 100 extracts “crab”, “Hokkaido”, “fresh”, and “shipping included” as words that can be a query. In addition, for the content C02, the search device 100 extracts “crab”, “brake”, “titanium”, and “shipping included” as words that can be a query.

  Subsequently, the search device 100 combines the extracted words as words that can be a query. Then, the search device 100 calculates a predetermined index value for each word based on the co-occurrence degree of the combined words (step S03). Here, the predetermined index value is a numerical value used when the search device 100 searches for content, and is a numerical value used for calculating a matching score with a query. Such an index value is set for each content and for each extracted word. Specifically, the index value according to the embodiment is the weight of the word “crab” included in the content C01 when the query is “crab” for the content C01. It has been digitized.

  The search device 100 obtains an index value of each word based on the co-occurrence degree of the combination of two words through a predetermined learning process, for example. For example, when “crab” as a query is transmitted from a predetermined user, the search device 100 clicks (selects) the word “crab” included in the predetermined content, which is displayed as a search result by the user. Learn the results of how much they contributed. In this case, the content clicked by the user indicates that the content selected by the user is considered to be content in line with his / her intention among a plurality of content displayed on the user terminal 10 as a search result. In other words, the clicked content (or a word included in the content) indicates correct answer data in the learning process.

  That is, when the content is searched for by the query “crab” and the searched content is clicked by the user, the search device 100 includes the word “query” and the word included in the clicked content. Determined to have co-occurrence. For example, if the content searched for by the query “crab” contains a high percentage of “crab”, “Hokkaido”, “fresh”, etc. in the content clicked by the user, <query “crab”, word “ "Crab">, <Query "Crab", word "Hokkaido">, and <Query "Crab", word "Fresh"> have co-occurrence between each word (co-occurrence) The degree is high). On the other hand, if the percentage of content searched for by the query “crab” includes “shipping included” in the content clicked by the user is low, the pair of <query “crab” and word “shipping included”> It is determined that the words do not have co-occurrence (low co-occurrence).

  In this way, the search device 100 contributes to how much the word used as the query contributes to the event of “searching for content clicked by the user (in other words, content intended by the user)”, This trend is quantified. More specifically, when a predetermined word (for example, “crab”) is used as a query in the search process, the search apparatus 100 uses a word (for example, “crab” or “crab” included in the content that is output as the search result. (Hokkaido ”,“ fresh ”,“ shipping included ”) and the word (“ crab ”) used as the query are determined to have a co-occurrence relationship, and the co-occurrence relationship is digitized based on learning.

  This will be specifically described with reference to the example of FIG. For the content C01 in the example of FIG. 1, the search device 100 extracts “crab”, “Hokkaido”, “fresh”, and “shipping included” as words that can be a query. At this time, when the query is “crab”, the search device 100 calculates a numerical value indicating the co-occurrence degree of each word by the machine learning. That is, the search device 100 calculates <index “crab”, word “crab”>, <query “crab”, word “Hokkaido”>, in order to calculate an index value corresponding to the word “crab” in the content C01. Four pairs are created: <query “crab”, word “fresh”>, <query “crab”, word “shipping included”>. And the search device 100 calculates the numerical value calculated | required by the said machine learning with respect to each of four pairs. Then, the search device 100 calculates a numerical value obtained by adding the numerical values of each of the four pairs as an index value of “the word“ crab ”in the content C01”.

  For example, the numerical value calculated for the pair <query “crab”, word “crab”> is “5”, and the numerical value calculated for the pair <query “crab”, word “Hokkaido”> “3”, the value calculated for the pair <query “crab”, word “fresh”> is “2” and for the pair <query “crab”, word “shipping included”> Assume that the calculated numerical value is “0”. In this case, the search device 100 calculates a numerical value “5 + 3 + 2 + 0” that is a numerical value obtained by adding the numerical values of each of the four pairs, that is, a numerical value “10” as an index value of “the word“ crab ”in the content C01”.

  Similarly, the search device 100 calculates an index value for words other than “crab” in the content C01. For example, since the search device 100 calculates an index value corresponding to the word “Hokkaido” in the content C01, <query “Hokkaido”, word “crab”>, <query “Hokkaido”, word “Hokkaido”>, < Four pairs are created: query “Hokkaido”, word “fresh”>, <query “Hokkaido”, word “shipping included”>. And the search device 100 calculates the numerical value calculated | required by the said machine learning with respect to each of four pairs. Then, the search device 100 calculates a numerical value obtained by adding the numerical values of each of the four pairs as an index value of “the word“ Hokkaido ”in the content C01”.

  In this way, the search device 100 calculates “10” for “crab”, “8” for “Hokkaido”, and “5” for “fresh” as the index value of each word in the content C01. It is assumed that “0.1” is calculated for “shipping included”. For example, when the query is “crab” in the content group to be searched by the search device 100, these index values include content that includes words such as “crab”, “Hokkaido”, and “fresh” in the title. This indicates that the user tends to be easily clicked. In other words, the user who transmitted the query “crab” has a tendency to be relatively easy to click on content including words such as “crab”, “Hokkaido”, and “fresh” in the title. That is, the query “crab” can be said to be a word that has a relatively large contribution to an event in which a user clicks on content including words such as “crab”, “Hokkaido”, and “fresh” in the title.

  On the other hand, in the example shown in FIG. 1, for example, when the query is “shipping included”, content that includes words such as “crab”, “Hokkaido”, and “fresh” in the title tends to be clicked by the user. It shows that it cannot be said. In other words, the user who sent the query “shipping included” does not intend to actively search for content that includes the words “crab”, “Hokkaido”, or “fresh” in the title by the query “shipping included”. It shows that. In other words, the query “shipping included” can be said to be a word that contributes little to an event in which content including words such as “crab”, “Hokkaido”, and “fresh” is clicked by the user. This means that just because a user sends a query “shipping included”, it cannot be said that the user is trying to search for content that includes words such as “crab”, “Hokkaido”, or “fresh” in the title. Is shown.

  Next, the content C02 will be described. For the content C02 in FIG. 1, the search device 100 extracts “crab”, “brake”, “titanium”, and “shipping included” as words that can be a query. When the query is “crab”, the search device 100 calculates a numerical value indicating the co-occurrence degree of each word by the machine learning. That is, the search device 100 calculates <index “crab”, word “crab”>, <query “crab”, word “brake”>, in order to calculate an index value corresponding to the word “crab” in the content C02. Four pairs are created: <query “crab”, word “titanium”>, <query “crab”, word “shipping included”>. And the search device 100 calculates the numerical value calculated | required by the said machine learning with respect to each of four pairs. Then, the search device 100 calculates a numerical value obtained by adding the numerical values of each of the four pairs as an index value of “the word“ crab ”in the content C02”.

  In the example of FIG. 1, as a tendency of the entire user who transmitted the query “crab”, “crab” or “crab” or “fresh” is compared with content including words such as “crab”, “Hokkaido”, or “fresh” in the title. It is assumed that a small number of users have made a search for content that includes words such as “brake” and “titanium” in the title. This means that a user who searches for content (products) such as “brake” using “crab” as a query compared to a user who searches for content (products) such as “crab” using “crab” as a query. Means very few.

  Therefore, a lower index value is calculated for “crab” in the content C02 than for “crab” in the content C01. Specifically, the numerical value calculated for the pair <query “crab”, word “crab”> is “5”, and calculated for the pair <query “crab”, word “brake”>. The numerical value calculated for the pair <query “crab”, word “titanium”> is “−5”, and <query “crab”, word “shipping included”> Assume that the numerical value calculated for the pair is “0”. In this case, the search device 100 sets “5 + (− 3) + (− 5) +0”, which is a numerical value obtained by adding the numerical values of each of the four pairs, that is, the numerical value “−3” to the “word in the content C02. Calculated as the index value for “crab”. Although FIG. 1 shows an example in which the index value takes a negative value, these numerical values indicate relative values, and the numerical values themselves do not make sense. For example, the search device 100 may adjust the values of the calculated index values to be all positive values by normalization.

  Similarly, the search device 100 calculates an index value for words other than “crab” in the content C02. That is, since the search device 100 calculates an index value corresponding to the word “brake” in the content C02, <query “brake”, word “crab”>, <query “brake”, word “brake”>, < Four pairs are created: query "brake", word "titanium">, <query "brake", word "shipping included">. And the search device 100 calculates the numerical value calculated | required by the said machine learning with respect to each of four pairs. Then, the search device 100 calculates a numerical value obtained by adding the numerical values of each of the four pairs as an index value of “the word“ brake ”in the content C02”.

  In this way, the search device 100 calculates “−3” for “crab”, “1” for “brake”, and “1” for “titanium” as the index value of each word in the content C02. And “0.2” is calculated for “shipping included”. For example, if the query is “crab” in the content group to be searched by the search device 100, the content including words such as “brake”, “titanium”, and “shipping included” in the title is “ Compared to content that includes words such as “crab”, “Hokkaido”, and “fresh” in the title, it indicates that it is less likely to be clicked by the user. That is, the query “crab” can be said to be a word that has a relatively small contribution to an event in which a user clicks on content including words such as “brake”, “titanium”, and “shipping included” in the title.

  As described above, the search device 100 calculates an index value in advance for the word C01 and the content C02 submitted by the exhibitor for a word that is assumed to be used for the search prior to the actual search process. Keep it. Then, the search device 100 stores the content in a predetermined storage unit in association with the calculated index value (step S04).

  Next, an actual search process performed by the search device 100 will be described with reference to FIG. FIG. 2 is a diagram (2) illustrating an example of a search process according to the embodiment. Specifically, FIG. 2 illustrates a process in which the search device 100 searches for content that matches the query transmitted by the user from the content held.

  A user terminal 10 shown in FIG. 2 is an information processing terminal used by a user. The user terminal 10 performs a process of displaying an auction site screen or transmitting a query input by the user on the auction site in accordance with a user operation.

  In FIG. 2, it is assumed that the user visits the auction site (step S10). Then, the user transmits two words “crab” and “shipping included” as a query (step S11). That is, the user uses the two words “crab” and “shipping included” as a query to the auction site. Search for exhibited items. Specifically, the user inputs the words “crab” and “shipping included” in the search window in the web page of the auction site displayed on the user terminal 10.

  The search device 100 matches the index value calculated in advance with the query for the stored content, and calculates a matching score (step S12). As described with reference to FIG. 1, the search device 100 has already calculated an index value of “crab: 10, Hokkaido: 8, fresh: 5, shipping included: 0.1” for the words included in the content C01. Yes. And the search device 100 totals the index value matched with the word matched with the word contained in a query among the words contained in a content. In this case, the search device 100 may use, for example, the following search model (the following formula (1)).

  In the above formula (1), “q” indicates “query”, and “D” indicates “document (in the embodiment, the document means content to be searched)”. , “T” indicates “a word (term) that exists in common in the query and document (content)”, w (t) indicates “term weight”, and f (t, D) indicates “Matching score” shall be indicated. In order to simplify the description, w (t) is assumed to be “1” in the embodiment.

  As indicated by the above formula (1), the search device 100 calculates the matching score between the query and the content by adding all the index values for the words common to the query (matched words). For example, for the content C01, the search device 100 calculates that “crab” and “shipping included” match, and the matching score is “10 + 0.1 = 10.1”. In addition, for the content C02, the search device 100 calculates that “crab” and “shipping included” match and the matching score is “−3 + 0.2 = −2.8”.

  Then, the search device 100 gives a rank to the content based on the matching score (step S13). For example, when the search target is the content C01 and the content C02, the search device 100 ranks the content C01 as “1” and ranks the content C02 as “2” based on the matching score.

  And the search device 100 transmits a search result with respect to the user terminal 10 in the rank order (step S14). That is, the search device 100 transmits the content C01 to the user terminal 10 with priority over the content C02 as a search result. The user terminal 10 displays the content transmitted from the search device 100 as a search result. Note that the user terminal 10 may transmit result information such as that the content displayed as the search result is clicked by the user or not clicked to the search device 100. Based on such information, the search device 100 further repeats a learning process for calculating an index value.

  As described with reference to FIGS. 1 and 2, the search device 100 according to the embodiment extracts features (words included in the content title in the example of FIG. 1) to be used for search processing from the content. In addition, the search device 100 calculates a predetermined index value set for each word based on the co-occurrence relationship in the extracted word combination, associates the calculated index value with the content, and stores the predetermined index value. Store in the department. Furthermore, the search device 100 searches for content that matches the query transmitted from the user based on the calculated predetermined index value.

  As a result, the search apparatus 100 includes the content C01 and the content C02 that include the same word (“crab” or “shipping included”) in the content title, and the query transmitted from the user is “crab” and “shipping included”. ”, It is possible to preferentially search for the content C01 that is likely to be assumed by the user. In addition, the search device 100 can perform a search process at high speed by extracting a word that can be a query from the content in advance and calculating an index value in advance for the extracted word. In addition, the search device 100 calculates an index value to be used for search processing in advance at the time of content index creation. In this case, the search device 100 uses a method of calculating an index value with a combination created only with words in the content without using a query transmitted from the user, thereby limiting the combinations of words to be processed. Can do. In this way, the search device 100 according to the embodiment can efficiently search for content according to the user's intention.

  As described above, the search device 100 increases the efficiency of search processing by calculating an index value at the time of content submission (at the time of index creation of content to be searched). Hereinafter, the configuration of the search device 100 that performs such processing and the search system 1 including the search device 100 will be described in detail.

[2. Search system configuration)
The configuration of the search system 1 including the search device 100 according to the embodiment will be described with reference to FIG. FIG. 3 is a diagram illustrating a configuration example of the search system 1 according to the embodiment. As illustrated in FIG. 3, the search system 1 according to the embodiment includes a user terminal 10, a submission terminal 20, a web server 30, and a search device 100. These various apparatuses are communicably connected via a network N (for example, the Internet) in a wired or wireless manner. Note that the search system 1 shown in FIG. 3 may include a plurality of user terminals 10, a plurality of submission terminals 20, and a plurality of web servers 30.

  The user terminal 10 is, for example, an information processing apparatus such as a smartphone, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a mobile phone, a PDA (Personal Digital Assistant), or a wearable device. It is. The user terminal 10 acquires the web page from the website provided from the web server 30 by accessing the web server 30 according to the operation by the user. Then, the user terminal 10 displays the acquired web page on a display device (for example, a liquid crystal display). In the present specification, the user and the user terminal 10 may be identified with each other. For example, “transmit content to the user” may actually mean “transmit content to the user terminal 10 used by the user”.

  The submission terminal 20 is an information processing device used by an exhibitor who submits content to the search device 100. The submission terminal 20 executes processing for submitting content to the search device 100 according to an operation by the exhibitor.

  In addition, an exhibitor may request an agency for such submission without submitting content to the search device 100 using the submission terminal 20. In this case, it is the agency that submits the content to the search device 100. In the following, the notation “exhibitor” is a concept that includes not only the seller but also the agency, and the expression “traffic terminal” includes the agency device used by the agency as well as the trafficking terminal. It is assumed to be a concept.

  The web server 30 is a server device that provides various web pages when accessed from the user terminal 10. In the embodiment, it is assumed that the web server 30 provides an auction site. The web server 30 provides various web pages related to, for example, news sites, weather forecast sites, shopping sites, finance (stock price) sites, route search sites, map providing sites, travel sites, restaurant introduction sites, web blogs, and the like. May be.

  The web page provided by the web server 30 includes a search window for searching the website. The user searches the content in the website by entering a query in the search window. In the embodiment, an example in which the process of providing a web page is performed by the web server 30 and the process of searching for content is performed by the search apparatus 100 is shown. However, the web server 30 and the search apparatus 100 are integrated. There may be.

  The search device 100 is a server device that extracts features included in the content received from the submission terminal 20, calculates an index value for the feature, and searches for the content based on the calculated index value.

[3. (Configuration of search device)
Next, the configuration of the search device 100 according to the embodiment will be described with reference to FIG. FIG. 4 is a diagram illustrating a configuration example of the search device 100 according to the embodiment. As illustrated in FIG. 4, the search device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The search device 100 includes an input unit (for example, a keyboard and a mouse) that receives various operations from an administrator who uses the search device 100, and a display unit (for example, a liquid crystal display) for displaying various types of information. You may have.

(About the communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card). The communication unit 110 is connected to the network N by wire or wirelessly, and transmits / receives information to / from the user terminal 10, the submission terminal 20, and the web server 30 via the network N.

(About the storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory device such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 includes a content storage unit 121, a search information storage unit 122, and an index value storage unit 123.

(About the content storage unit 121)
The content storage unit 121 stores information related to content submitted from the submission terminal 20. Here, FIG. 5 shows an example of the content storage unit 121 according to the embodiment. FIG. 5 is a diagram illustrating an example of the content storage unit 121 according to the embodiment. In the example illustrated in FIG. 5, the content storage unit 121 includes items such as “exhibitor ID”, “content ID”, “title”, “product”, “price range”, and “product category”.

  “Exhibitor ID” indicates identification information for identifying the exhibitor or the submission terminal 20. “Content ID” indicates identification information for identifying content submitted by the exhibitor. In this specification, identification information as shown in FIG. 5 may be used as a reference symbol. For example, the exhibitor identified by the exhibitor ID “A01” may be referred to as “exhibitor A01”, and the content identified by the content ID “C01” may be denoted as “content C01”.

  “Title” indicates a title set in the content. “Product” indicates a product (exhibit) associated with the content. “Price range” indicates the price range to which the price set for the product by the exhibitor belongs. For example, a product whose exhibition price set for the product by the seller is “1500” yen belongs to the price range of “1000 to 2000” yen. “Product category” indicates the category to which the product belongs.

  That is, the example of the data shown in FIG. 5 indicates that the seller A01 identified by the seller ID “A01” has submitted the content C01 identified by the content ID “C01”. The title of content C01 is “I will deliver fresh crab from Hokkaido (including shipping!)”, The product is “crab”, the price range is “1000-2000” yen, and the product category is It shows that it is “fresh food”.

  Although not shown in FIG. 5, the content storage unit 121 may store information such as a description of a product set in the content, image data obtained by capturing the product, and date and time of submission. .

  Further, content data (text data, image data, etc.) of content that is actually transmitted to the user terminal 10 may be stored in a predetermined storage server provided separately from the search device 100. In this case, the search device 100 specifies the content stored in the external storage server based on the content ID stored in the content storage unit 121. Then, the search device 100 controls the storage server to transmit the specified content to the user terminal 10.

(Search information storage unit 122)
The search information storage unit 122 stores information related to content searched by the user. Specifically, the search information storage unit 122 stores information (click log) on the user's reaction to the content transmitted as the search result. Here, FIG. 6 illustrates an example of the search information storage unit 122 according to the embodiment. FIG. 6 is a diagram illustrating an example of the search information storage unit 122 according to the embodiment. In the example illustrated in FIG. 6, the search information storage unit 122 includes items such as “search ID”, “search query”, “clicked content ID”, and “word included in content”.

  “Search ID” indicates identification information for identifying the executed search process. The “search query” indicates a query transmitted from the user in the search process. “Clicked content ID” is the identification of the content actually clicked by the user among the plurality of contents (for example, the plurality of contents displayed in a list on the search result page) transmitted to the user terminal 10 as the search result. Indicates information. It is assumed that the content ID shown in FIG. 6 matches the content ID shown in FIG. “Word included in content” indicates a word included in the title of the content clicked by the user. Note that the word included in the content in this case corresponds to the word extracted as the “word that can be a query” included in the content shown in FIG.

  6 is log data of the search F01 identified by the search ID “F01”, the search query is “fresh” and “crab”, and the clicked content ID is “C01”. ”And the word included in the content C01 indicates“ crab ”,“ Hokkaido ”,“ fresh ”,“ shipping included ”.

(About the index value storage unit 123)
The index value storage unit 123 stores the index value calculated by the search device 100 and the content in association with each other. Here, FIG. 7 shows an example of the index value storage unit 123 according to the embodiment. FIG. 7 is a diagram illustrating an example of the index value storage unit 123 according to the embodiment. In the example illustrated in FIG. 7, the index value storage unit 123 includes items such as “content ID”, “extracted word”, and “index value”.

  The “content ID” corresponds to the same item shown in FIGS. The “extracted word” indicates a word extracted as “a word that can be a query” among the words included in the title of the content. The “index value” indicates an index value calculated for each extracted word in each content.

  That is, in the example of the data shown in FIG. 7, in the content C01 identified by the content ID “C01”, the extracted words are “crab”, “Hokkaido”, “fresh”, “shipping included”, The index values calculated for each of these are “10”, “8”, “5”, and “0.1”.

(About the control unit 130)
The control unit 130 is a controller, for example, various programs (an example of a search program) stored in a storage device inside the search device 100 by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. This is realized by executing the RAM as a work area. The control unit 130 is a controller, and is realized by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  As shown in FIG. 4, the control unit 130 includes a submission receiving unit 131, an extraction unit 132, an acquisition unit 133, a calculation unit 134, a reception unit 135, a search control unit 136, a generation unit 137, It has a search unit 138 and a transmission unit 139, and realizes or executes information processing functions and operations described below. Note that the internal configuration of the control unit 130 is not limited to the configuration illustrated in FIG. 4, and may be another configuration as long as the information processing described below is performed. In addition, the connection relationship between the processing units included in the control unit 130 is not limited to the connection relationship illustrated in FIG. 4, and may be another connection relationship.

(About the submission acceptance part 131)
The submission accepting unit 131 accepts content submission from the submission terminal 20. Then, the submission receiving unit 131 stores the seller ID for identifying the exhibitor as the submission source in association with the content ID for identifying the content in the content storage unit 121. Further, the submission accepting unit 131 may accept the setting of the exhibition price for the product associated with the content, the setting of the product category, and the like from the submission terminal 20.

(About the extraction unit 132)
The extraction unit 132 extracts features to be used for search processing from the content. For example, the extraction unit 132 extracts words included in the content as the feature. In the embodiment, the extraction unit 132 shows an example of extracting a word that can be a query from words included in the title of the content.

  For example, the extraction unit 132 performs morphological analysis on text data constituting the title of the content. And the extraction part 132 extracts the word with many frequency used as a query in the past among the words which carried out the morphological analysis as a word which can become a query in the said content.

  Note that the extraction unit 132 may extract a word that can be a query using a weight value such as tf-idf. For example, the extraction unit 132 is a word that is widely used in general in the past even if it is a word that has been frequently used as a query, and is determined not to be a word indicating the characteristics of the content (for example, , A word having a low weight value) may not be extracted.

(About the acquisition unit 133)
The acquisition unit 133 acquires various types of information. For example, the acquisition unit 133 acquires information regarding the user's reaction in the search process, such as which content is clicked on the user terminal 10 that has received the content search result. For example, every time content is transmitted to the user terminal 10 by the transmission unit 139 described later, the acquisition unit 133 acquires information regarding a user reaction such as which content is clicked on the user terminal 10. Then, the acquisition unit 133 stores the acquired information in the search information storage unit 122 as log data related to the search process.

  In addition, after the feature of the content is extracted by the extraction unit 132, the acquisition unit 133 acquires the search log acquired so far from the search information storage unit 122 for use in the index value calculation process for the content. . For example, as illustrated in FIG. 6, the acquisition unit 133 displays information regarding what kind of content is actually clicked by the user when the query is searched, and information on a word included in the clicked content. get. That is, the acquisition unit 133 acquires information related to the query in the search process, the number of times the content searched by the query is clicked, the number of appearances of words included in the clicked content, and the like.

  Note that the acquisition unit 133 includes information on a plurality of contents included in the search result page displayed on the user terminal 10 (eg, up to which rank the content has been displayed on the search result page), and the content on the user terminal 10. The method for acquiring information such as whether or not the button has been clicked may be realized by various known methods. For example, the acquisition unit 133 may acquire information on content and operation information on the user terminal 10 using a notification function realized by a web beacon or the like.

(About the calculation unit 134)
The calculation unit 134 calculates a predetermined index value set for each feature based on the co-occurrence relationship in the combination of features extracted by the extraction unit 132, associates the calculated index value with the content, and sets the predetermined index value. Stored in the storage unit.

  Specifically, the calculation unit 134 quantifies the co-occurrence relationship in the word combination extracted by the extraction unit 132 based on the search information acquired by the acquisition unit 133. Then, the calculation unit 134 calculates an index value set for each word in the content based on the digitized co-occurrence relationship.

  As described with reference to FIG. 1, when a word is used as a query in the search process, the calculation unit 134 calculates a word included in the content output as the search result in the search process and a word used as the query. It is determined that there is a co-occurrence relationship.

  Then, when the predetermined word is used as a query in the search process, the calculation unit 134 learns a result indicating whether or not the content desired by the user who transmitted the query is searched as a search result. A predetermined index value set for the word is calculated.

  For example, the calculation unit 134 learns that the predetermined index value set for the word used as the query in the content increases as the number of times the content is selected as the search result from the user who transmitted the query increases. Based on the result, a predetermined index value is calculated.

  As a specific flow of calculation processing, first, the calculation unit 134 learns information about the search result acquired by the acquisition unit 133, so that the word used as the query “how much the content clicked by the user is determined. The tendency of whether or not it contributed to the event “search” is quantified. Then, the calculation unit 134 determines that the word as the query in the search process and the word included in the content output as the search result have a co-occurrence relationship. For example, the calculation unit 134 increases the number of combinations of the word as a query and the word included in the content searched by the query from among the search results having a certain number of samples. A high numerical value is calculated assuming that the degree of co-occurrence between them is high. Note that the process of digitizing the co-occurrence relationship between words may be performed by variously combining known machine learning processes.

  Then, the calculation unit 134 creates a pair with the words extracted in the content, and calculates an index value for each word in the content. In the example of the content C01 illustrated in FIG. 1, the calculation unit 134 uses the numerical value “5” of the pair <query “crab” and word “crab”> as the index value of the word “crab” in the content C01, and <query The number “3” for the pair “crab” and the word “Hokkaido”>, the number “2” for the pair “query“ crab ”and the word“ fresh ””, and <query “crab” and the word “shipping included”> All the numerical values “0” of the pair are added to calculate “10”.

  In this way, the calculation unit 134 is an index for search processing in a stage before performing a search using an actually transmitted query, that is, in a stage of indexing content to be searched in the storage unit. Calculate the value. Then, the calculation unit 134 associates the calculated index value with the content and stores them in the index value storage unit 123.

(Receiver 135)
The receiving unit 135 receives information related to the query transmitted from the user terminal 10. Specifically, the reception unit 135 receives a search query that is transmitted from the user terminal 10 that displays a web page and is used to search for content included in the web page. For example, the receiving unit 135 receives a word or a combination of words as a query. The receiving unit 135 may receive the price set for the product associated with the content, the category of the product, and the like as a query.

(About the search control unit 136)
The search control unit 136 controls processing related to content search. In the embodiment, the search control unit 136 uses the index value calculated by the calculation unit 134 to generate a model for performing a search, and the content generated using the model generated by the generation unit 137. A search unit 138 for searching.

(About the generation unit 137)
Based on the index value calculated by the calculation unit 134, the generation unit 137 generates a model for searching for content that matches the query from the search target content. Specifically, as shown in the above formula (1), when the word included in the content matches the query, the generating unit 137 calculates the index value by summing the index values of the matched words. A simple search model.

  Note that the model expressed by the above formula (1) is an example, and when the word included in the content matches the query, the generation unit 137 calculates the matching score in the content based on the index value of the matched word. If it is a model to be output, a model different from the above equation (1) may be generated.

(About the search unit 138)
The search unit 138 searches for content that matches the query transmitted from the user, based on the predetermined index value calculated by the calculation unit 134. Specifically, the search unit 138 inputs the query transmitted from the user and the search target content to the model generated by the generation unit 137, and based on the matching score output from the model according to the input, Search for content that matches the query.

  Specifically, the search unit 138 ranks the content in the order of the output matching score. And the search part 138 arranges the content as a search result so that a rank may be displayed as a search result in an order from a high-order content. That is, the content having a higher matching score by the search unit 138 is treated as a higher content in the search result.

(About transmitter 139)
The transmission unit 139 transmits various information. For example, in response to the query received by the reception unit 135, the transmission unit 139 transmits the content searched by the search unit 138 to the user terminal 10 as a search result.

  Note that, as described above, content data itself that is actually transmitted to the user terminal 10 may not be stored in the content storage unit 121 of the search device 100. For example, the transmission unit 139 may cause the content to be transmitted to the user terminal 10 by transmitting a control command to a predetermined storage server provided outside.

[4. Web server configuration]
Next, the configuration of the web server 30 according to the embodiment will be described with reference to FIG. FIG. 8 is a diagram illustrating a configuration example of the web server 30 according to the embodiment. As illustrated in FIG. 8, the web server 30 includes a communication unit 31, a web page storage unit 32, and a control unit 33.

  The communication part 31 is implement | achieved by NIC etc., for example. The communication unit 31 is connected to the network N by wire or wireless, and transmits / receives information to / from the user terminal 10 or the search device 100.

  The web page storage unit 32 is realized by, for example, a semiconductor memory device such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. And the web page memory | storage part 32 memorize | stores the information regarding a web page. For example, the web page storage unit 32 stores an HTML file forming a web page, a still image or a moving image displayed on the web page. The web page stored in the web page storage unit 32 includes a search window for searching for content on the web page and a search frame for searching for content (products) by setting a price range and category. May be.

  The control unit 33 is realized, for example, by executing various programs stored in a storage device inside the web server 30 using the RAM as a work area by a CPU, an MPU, or the like. The control unit 33 is realized by an integrated circuit such as an ASIC or FPGA, for example.

  As illustrated in FIG. 8, the control unit 33 includes a reception unit 34 and a distribution unit 35, and realizes or executes information processing functions and operations described below. The internal configuration of the control unit 33 is not limited to the configuration illustrated in FIG. 8, and may be another configuration as long as the information processing described later is performed. In addition, the connection relationship between the processing units included in the control unit 33 is not limited to the connection relationship illustrated in FIG. 8 and may be another connection relationship.

  The accepting unit 34 accepts a web page acquisition request from the user terminal 10. For example, the reception unit 34 receives an HTTP request as a web page acquisition request.

  The distribution unit 35 distributes the web page to the user terminal 10 when the reception unit 34 receives a web page acquisition request. Specifically, the distribution unit 35 acquires an acquisition request target web page from the web page storage unit 32 and distributes the acquired web page to the user terminal 10. In addition, when receiving a search query input from the user terminal 10, the distribution unit 35 may transmit the input content to the search device 100. Alternatively, when receiving a search query input from the user terminal 10, the distribution unit 35 controls communication between the user terminal 10 and the search device 100 so that the search query is transmitted from the user terminal 10 to the search device 100. You may control.

[5. Processing procedure)
Next, a processing procedure performed by the search device 100 according to the embodiment will be described with reference to FIGS. 9 and 10. First, referring to FIG. 9, a description will be given of a processing procedure in which the search device 100 calculates an index value when content is submitted, and stores the content and the index value in association with each other in the storage unit. FIG. 9 is a flowchart (1) illustrating a processing procedure according to the embodiment.

  As shown in FIG. 9, the search apparatus 100 determines whether or not content submission has been received from the submission terminal 20 (step S101). When content submission is not received (step S101; No), the search device 100 stands by until content submission is received.

  On the other hand, when the content submission is received (step S101; Yes), the search device 100 extracts a feature (word) that can be a query from the title of the content (step S102).

  And the search device 100 acquires the search information regarding the extracted feature (step S103). For example, when the extracted feature is used as a query as the search information, the search device 100 includes what content is searched, what content is clicked, and what word is clicked. Get information such as if it was.

  Subsequently, the search device 100 calculates an index value related to the feature based on the acquired search information (step S104). Then, the search device 100 associates the content with the index value calculated for each feature in the content and stores it in the index value storage unit 123 (step S105).

  Next, the procedure of the process in which the search device 100 searches for content will be described with reference to FIG. FIG. 10 is a flowchart (2) illustrating a processing procedure according to the embodiment.

  As illustrated in FIG. 10, the search device 100 determines whether a query is received from the user terminal 10 (step S201). If no query has been received (step S201; No), the search device 100 waits until a query is received.

  On the other hand, when a query is received (step S201; Yes), the search device 100 calculates a matching score based on the query and the index value (step S202). Then, the search device 100 gives a rank to the content based on the matching score (step S203).

  Subsequently, the search device 100 transmits the contents arranged in the rank order to the user terminal 10 as a search result (step S204). Further, the search device 100 acquires information related to the user's reaction to the content (step S205). Specifically, the search device 100 acquires information regarding which of the contents transmitted as a search result has been clicked as information related to the user's reaction.

  Then, after acquiring information related to the user's reaction to the content, the search device 100 performs learning processing based on the acquired information and updates the information related to the content (step S206). For example, the search device 100 recalculates numerical values related to word co-occurrence relationships and index values of words in content, and optimizes index values.

[6. (Modification)
The search device 100 described above may be implemented in various different forms other than the above embodiment. Therefore, in the following, another embodiment of the search device 100 will be described.

[6-1. (Word weight)
In the above embodiment, an example in which the value of the word weight (w (t)) is “1” in the model generated by the search device 100 has been described. However, the search device 100 may use a numerical value other than “1” as the word weight. For example, the search device 100 may give a weight as a word to a word that tends to be frequently used in a query.

  For example, the search device 100 refers to the search information storage unit 122 and sets a word with a higher weight as a word used more frequently as a query among words of a query transmitted from the user. For example, it is assumed that the number of times each of the two words “crab” and “shipping included” is used as a query is “9: 1”. In this case, when the user transmits a combination of two words “crab” and “shipping included” as a query, the search apparatus 100 sets “0.9” to “crab” as the weight of the word, “0.1” may be set in “shipping included”.

  In this case, taking the content C01 shown in FIGS. 1 and 2 as an example, in calculating the matching score, the search apparatus 100 matches “crab” and “shipping included”, and the matching score is “10 * 0. 9 + 0.1 * 0.1 = 9.01 ”. Taking the content C02 as an example, in calculating the matching score, the search apparatus 100 matches “crab” and “shipping included”, and the matching score is “−3 * 0.9 + 0.2 * 0.1 = -2.68 ". That is, the search device 100 gives weight values to words (queries) that are more likely to be used as queries to the user, so that the matching score calculated for words that are frequently used as queries is less frequently used as queries. Adjustment can be made so that it is calculated higher than the matching score calculated for the word. Thereby, the search device 100 can further reflect the user's intention in the search result.

[6-2. (Feature information)
In the embodiment described above, the search apparatus 100 has shown an example in which the index value calculation process and the search process using the index value are performed using the word included in the content title as a feature. Here, the search device 100 may use information other than words as features.

  For example, the search device 100 may extract feature information set for a product associated with the content as the feature. Then, the search device 100 may calculate an index value set for each word or feature information based on the co-occurrence relationship in any combination of extracted words or feature information. For example, the search device 100 extracts a product price or a product category as the feature information. In this case, the search device 100 calculates an index value set for each word, product price, or product category based on the co-occurrence relationship in any combination of the extracted word, product price, or product category. calculate.

  Taking FIG. 5 as an example, the search device 100 has a price range of “1000-2000” yen as feature information set for the product “crab” associated with the content C01, and the product category is “fresh food”. Therefore, the feature information is used as a feature. That is, in an auction site or the like, a user may search for a product by price or category as well as words input to the search window as a query. In this case, the search device 100 receives a price and a category as a query. And the search device 100 acquires the value which digitized the co-occurrence information of the paired information among information, such as a word, a price, and a category similarly to the process demonstrated by the said embodiment. Thereby, the search device 100 can calculate index values for not only words but also prices, categories, and the like.

  Then, the search device 100 stores the content in association with the index value calculated for each word, price, category, and the like. When the user transmits a price or category as a query, matching between the query and the content feature (that is, characteristic information such as a word or price or category) is determined, and a matching score is calculated. As a result, the search device 100 can transmit, as a search result, content that better matches the query in response to a user who transmits a query by combining words, prices, and categories.

[6-3. (Feature)
Further, the search device 100 may use not only the word included in the title of the content and the price and category of the product associated with the content but also various information as the content features. For example, the search device 100 may extract words from all text data included in the content as well as the title of the content. Further, the search device 100 uses various information such as information on the size and weight of the product, the number of days and time until the end of the auction, color, shape, and pattern as features as the feature information of the product associated with the content. May be.

[6-4. (Matching score)
In addition, when calculating the index value or the matching score, the search device 100 may perform calculation in consideration of the frequency of features (words) included in the content. For example, the search device 100 may perform adjustment such as giving a weight to the word according to the frequency of the word appearing in the title of the content. For example, in the content C01 shown in FIG. 1, among the words included in the content title, “Hokkaido” appears twice as many as other words. In this case, the search device 100 considers the frequency and calculates, for example, the numerical value calculated for the pair <query “crab”, word “Hokkaido”> as “6”, which is twice “3”. Such adjustments may be made. Note that such a numerical value is an example, and the search device 100 may appropriately adjust the numerical value by, for example, normalizing the frequency of words applied to the content or indicating the frequency as a logarithm.

[6-5. (Medium)
In the above-described embodiment, an example in which search processing is performed using a web page provided by the web server 30 as an example has been described. However, a medium for displaying content is not limited to a web page such as an auction site. For example, the search device 100 may use the processing method described above for processing for searching for content in a service provided via a predetermined application, processing for searching for a predetermined document (document), and the like.

[6-6. Learning)
In the above-described embodiment, the example in which the search device 100 performs the index value calculation process by learning the tendency that the content transmitted as the search result is clicked by the user has been described. However, the search device 100 has not only a tendency to click, but also a tendency that the user has browsed the content, or a tendency that the user has converted (conversion) such as purchasing a product associated with the content. You may learn.

  Further, the search device 100 may search for content by generating a word vector corresponding to the content from words included in the content using the learning result and using a vector space. For example, the search device 100 calculates the spatial relationship between the query transmitted from the user and the word vector corresponding to the content, and searches the content corresponding to the word vector having a spatial direction close to the spatial direction indicated by the query. May be transmitted to the user.

[7. Hardware configuration)
The search device 100, the user terminal 10, and the web server 30 according to the above-described embodiments are realized by a computer 1000 having a configuration as shown in FIG. 11, for example. Hereinafter, the search device 100 will be described as an example. FIG. 11 is a hardware configuration diagram illustrating an example of a computer 1000 that implements the functions of the search device 100. The computer 1000 includes a CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I / F) 1500, input / output interface (I / F) 1600, and media interface (I / F) 1700.

  The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400 and controls each unit. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started up, a program depending on the hardware of the computer 1000, and the like.

  The HDD 1400 stores a program executed by the CPU 1100, data used by the program, and the like. The communication interface 1500 receives data from other devices via the communication network 500 (corresponding to the network N shown in FIG. 3), sends the data to the CPU 1100, and transmits the data generated by the CPU 1100 to the other devices via the communication network 500. Send to device.

  The CPU 1100 controls an output device such as a display and a printer and an input device such as a keyboard and a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. Further, the CPU 1100 outputs the data generated via the input / output interface 1600 to the output device.

  The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads such a program from the recording medium 1800 onto the RAM 1200 via the media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase change rewritable disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. Etc.

  For example, when the computer 1000 functions as the search device 100 according to the embodiment, the CPU 1100 of the computer 1000 implements the function of the control unit 130 by executing a program loaded on the RAM 1200. The HDD 1400 stores data in the storage unit 120. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be acquired from other devices via the communication network 500.

[8. Others]
In addition, among the processes described in the above embodiment, all or part of the processes described as being automatically performed can be performed manually, or the processes described as being performed manually can be performed. All or a part can be automatically performed by a known method. In addition, the processing procedures, specific names, and information including various data and parameters shown in the document and drawings can be arbitrarily changed unless otherwise specified. For example, the various types of information illustrated in each drawing is not limited to the illustrated information.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. For example, the acquisition unit 133 and the reception unit 135 illustrated in FIG. 4 may be integrated. Further, for example, information stored in the storage unit 120 may be stored in a predetermined storage device provided outside via the network N.

  Further, in the above-described embodiment, an example has been described in which the search device 100 performs, for example, a reception process that accepts content submission, a calculation process that calculates an index value, and a search process that searches for content. However, the search device 100 described above may be separated into a reception device that performs reception processing, a calculation device that performs calculation processing, and a search device that performs search processing. In this case, the reception device has at least a submission reception unit 131. The calculation device has at least a calculation unit 134. The search device has at least a search unit 138. And the process by said search device 100 is implement | achieved by the search system 1 which has each apparatus of a reception apparatus, a calculation apparatus, and a search apparatus.

  In addition, the above-described embodiments and modifications can be combined as appropriate within a range that does not contradict processing contents.

[9. effect〕
As described above, the search device 100 according to the embodiment includes the extraction unit 132, the calculation unit 134, and the search unit 138. The extraction unit 132 extracts features to be used for search processing from the content. The calculation unit 134 calculates a predetermined index value set for each feature based on the co-occurrence relationship in the combination of features extracted by the extraction unit 132, associates the calculated index value with the content, and sets the predetermined index value. In the storage unit (in the embodiment, the index value storage unit 123). The search unit 138 searches for content that matches the query transmitted from the user, based on the predetermined index value calculated by the calculation unit 134.

  As described above, the search device 100 according to the embodiment does not calculate the index value used for the search process when the query is transmitted from the user, but calculates the index value used for the search process at the time of storing the content. . Further, the search device 100 calculates an index value based on a co-occurrence relationship in a combination of content features. Thereby, the search device 100 can search efficiently the content according to the user's intention.

  In addition, the search device 100 according to the embodiment further includes a generation unit 137 that generates a model for searching for content that matches the query from the search target content based on a predetermined index value. The search unit 138 inputs the query transmitted from the user and the search target content to the model generated by the generation unit 137, and matches the query based on the matching score output from the model in response to the input. Search for.

  Thus, the search device 100 according to the embodiment performs a search using a search model that outputs a matching score based on the index value. In addition, as described above, the search device 100 calculates the index value for each feature included in the content in advance. Thereby, when the search device 100 receives a query from a user, the search device 100 can output a matching score quickly, and thus the search process can be speeded up.

  In addition, when a predetermined feature is used as a query in the search process, the calculation unit 134 establishes a co-occurrence relationship between the feature included in the content output as the search result in the search process and the feature used as the query. It is determined that it has.

  Thus, the search device 100 according to the embodiment determines that the feature used as the query and the feature included in the content output as the search result have a co-occurrence relationship. Since the content searched by a certain feature can be assumed to have a relationship with the feature, the search device 100 can appropriately quantify the co-occurrence relationship between the features extracted in the content by the above processing. it can.

  Further, when the feature is used as a query in the search process, the calculation unit 134 learns a result indicating whether or not the content desired by the user who transmitted the query has been searched as a search result. A predetermined index value to be set is calculated.

  As described above, the search device 100 according to the embodiment acquires the user's reaction to the result of the actual content search, and determines the co-occurrence relationship between the features based on the acquired information. The index value set for the feature is calculated. Therefore, the search device 100 can calculate an index value reflecting the user's reaction, that is, the user's intention indicated by the query, with respect to the index value. As a result, the search device 100 can search for content more in line with the user's intention.

  In addition, the calculation unit 134 learns that the predetermined index value set for the feature used as the query in the content increases as the number of times the content is selected as the search result from the user who transmitted the query increases. Based on the result, a predetermined index value is calculated.

  As described above, the search device 100 according to the embodiment acquires the number of times content is selected (for example, the number of clicks) as the user's response to the result of the actual content search. When the content intended by the user is included in the search result, it is assumed that the user terminal 10 displays detailed information or the like by clicking the content. For this reason, the search device 100 can calculate the index value that reflects the user's intention with high accuracy by calculating the index value based on the number of times content is selected as a search result from the user.

  Moreover, the extraction part 132 extracts the word contained in a content as a feature. The calculation unit 134 calculates a predetermined index value set for each word based on the co-occurrence relationship in the word combination extracted by the extraction unit 132.

  Thus, the search device 100 according to the embodiment performs processing using the words included in the content. As a result, the search device 100 more accurately searches for content that is likely to be assumed by the user in the query even when the search target is a plurality of content that includes the same word in the content. can do.

  Moreover, the extraction part 132 extracts the feature information set to the goods linked | related with the content as a feature. The calculation unit 134 calculates a predetermined index value set for each word or feature information based on the co-occurrence relationship in any combination of the words or feature information extracted by the extraction unit 132.

  As described above, the search device 100 according to the embodiment may perform processing using not only the word but also the feature information of the product included in the content. Thereby, the search device 100 can search the user for appropriate content according to the intention even in a situation where various information is used as a query, not limited to words.

  Further, the extraction unit 132 extracts a product price or a product category as the feature information. The calculation unit 134 is a predetermined unit set for each word, product price, or product category based on the co-occurrence relationship in any combination of the word, product price, or product category extracted by the extraction unit 132. An index value is calculated.

  Thus, the search device 100 according to the embodiment may perform processing using information such as the price and category of the product included in the content. Thereby, the search device 100 can provide the user with a search result that more appropriately reflects the user's intention in the search of an auction site, a shopping site, or the like that performs a search by specifying a price or a category.

  The embodiment of the present application has been described in detail with reference to the drawings. However, this is an exemplification, and various modifications and improvements are made based on the knowledge of those skilled in the art including the aspects described in the column of the disclosure of the invention. The present invention can be implemented in other forms.

  In addition, the “section (module, unit)” described above can be read as “means” or “circuit”. For example, the acquisition unit can be read as acquisition means or an acquisition circuit.

DESCRIPTION OF SYMBOLS 1 Search system 10 User terminal 20 Submission terminal 30 Web server 100 Search apparatus 110 Communication part 120 Storage part 121 Content storage part 122 Search information storage part 123 Index value storage part 130 Control part 131 Submission reception part 132 Extraction part 133 Acquisition part 134 Calculation Unit 135 Reception Unit 136 Search Control Unit 137 Generation Unit 138 Search Unit 139 Transmission Unit

Claims (10)

An extraction unit that extracts features to be used for search processing from the content;
Based on the co-occurrence relationship in the combination of features extracted by the extraction unit, a predetermined index value set for each of the features is calculated, and the calculated index value and the content are associated with each other and a predetermined storage unit A calculation unit to be stored in
A search unit that searches for content that matches a query transmitted from the user based on the predetermined index value calculated by the calculation unit;
A search device comprising: A generating unit that generates a model for searching for content that matches a query from search target content based on the predetermined index value;
Further comprising
The search unit
The query transmitted from the user and the search target content are input to the model generated by the generation unit, and content matching the query is obtained based on the matching score output from the model in response to the input. Search for,
The search device according to claim 1. The calculation unit includes:
When a predetermined feature is used as a query in the search process, it is determined that the feature included in the content output as the search result in the search process and the feature used as the query have a co-occurrence relationship.
The search device according to claim 1 or 2, wherein The calculation unit includes:
When the feature is used as a query in the search process, a result indicating whether or not the content desired by the user who transmitted the query has been searched is learned as a search result. Calculate the index value,
The search device according to any one of claims 1 to 3, wherein The calculation unit includes:
Based on the result of processing learned that the predetermined index value set for the feature used as the query in the content increases as the number of times the content is selected as the search result from the user who transmitted the query increases. Calculating the predetermined index value,
The search device according to claim 4, wherein: The extraction unit includes:
As the feature, a word included in the content is extracted,
The calculation unit includes:
Calculating a predetermined index value set for each word based on the co-occurrence relationship in the combination of words extracted by the extraction unit;
The search device according to any one of claims 1 to 5, wherein The extraction unit includes:
As the feature, feature information set for the product associated with the content is extracted,
The calculation unit includes:
Based on the co-occurrence relationship in any combination of the word or the feature information extracted by the extraction unit, a predetermined index value set for each word or the feature information is calculated.
The search device according to claim 6. The extraction unit includes:
As the feature information, extract the price of the product or the category of the product,
The calculation unit includes:
A predetermined value set for each of the word, the price of the product, or the category of the product based on a co-occurrence relationship in any combination of the word, the price of the product, or the category of the product extracted by the extraction unit. Calculate the index value of
The search device according to claim 7. A search method executed by a computer,
An extraction process for extracting features for use in search processing from the content;
Based on the co-occurrence relationship in the feature combination extracted by the extraction step, a predetermined index value set for each of the features is calculated, and the calculated index value and the content are associated with each other and a predetermined storage unit A calculation process to be stored in
A search step of searching for content that matches a query transmitted from the user based on the predetermined index value calculated by the calculation step;
The search method characterized by including. An extraction procedure for extracting features for use in search processing from the content;
Based on the co-occurrence relationship in the combination of features extracted by the extraction procedure, a predetermined index value set for each of the features is calculated, and the calculated index value and the content are associated with each other and a predetermined storage unit The calculation procedure to store in
A search procedure for searching for content that matches a query transmitted from the user based on the predetermined index value calculated by the calculation procedure;
A search program characterized by causing a computer to execute.

Images