JP2019053681A - Information processing apparatus, information processing method, and program - Google Patents

Abstract

To provide an information processing apparatus for improving search accuracy by estimating category accurately, an information processing method, and a program.SOLUTION: An information processing apparatus includes: an acquisition unit which acquires a query input by a user; a determination unit which determines whether the acquired query is included in category dictionary data formed by representing relationship between queries in a tree structure and associating each of the queries with categories to be searched; a first estimation unit which estimates a category corresponding to the acquired query, on the basis of the category dictionary data, when the determination unit determines that the acquired query is included in the category dictionary data; and a second estimation unit which estimates a category corresponding to the acquired query, on the basis of a model obtained by learning relationship between a query used for search and category, when the determination unit determines that the acquired query is not included in the category dictionary data.SELECTED DRAWING: Figure 9

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program.

  In the field of web search, in accordance with a keyword input by a user (hereinafter referred to as “query”), a method of providing information on various products or services related to the query (hereinafter referred to as “product etc.”) is performed. ing. In addition, in order to provide information such as products intended by the user with high accuracy, categories such as products corresponding to the query are defined in advance as data of a tree structure between categories, and based on the data of this tree structure, A method for calculating the similarity between categories for a query is known (see, for example, Patent Document 1).

JP 2017-45196

  However, in the conventional method, there is a case where information such as a product that is not intended by the user is provided because the degree of association with the query is low. For example, when the information on the product “TV stand” is provided for the query “TV”, the information about the product “glass” is provided for the query “beer”, etc. In some cases, information on a product belonging to a category different from the product or the like is provided.

  The present invention has been made in view of such circumstances, and provides an information processing apparatus, an information processing method, and a program capable of improving search accuracy by more accurately estimating a category. One of the purposes.

  According to an aspect of the present invention, an acquisition unit that acquires a query input by a user, a query acquired by the acquisition unit, a relationship between queries is represented in a tree structure, and each of the queries and a search target A determination unit that determines whether the category is associated with the category dictionary data associated with the category dictionary, and the determination unit determines that the query acquired by the acquisition unit exists in the category dictionary data, the category dictionary A first estimation unit that estimates a category corresponding to the query acquired by the acquisition unit based on data, and the determination unit determines that the query acquired by the acquisition unit does not exist in the category dictionary data A query acquired by the acquisition unit based on a model used to learn a relationship between a query used for search and a category; An information processing apparatus comprising: a second estimation unit that estimates a corresponding category; and a generation unit that generates information for displaying a search result belonging to the category estimated by the first estimation unit or the second estimation unit. .

  According to one embodiment of the present invention, search accuracy can be improved by accurately estimating a category.

It is a functional block diagram of search server 1 of a 1st embodiment. It is a figure which shows an example of the category tree data D2 of 1st Embodiment. It is a figure which shows an example of the word tree data D3 of 1st Embodiment. It is a flowchart which shows an example of the flow of the search process of the search server 1 of 1st Embodiment. It is a figure explaining the outline | summary of an example of the search process of the search server 1 of 1st Embodiment. It is a figure which shows an example of the relationship between the query processed by the search server 1 of 1st Embodiment, and the estimated category. It is a figure which shows an example of the category tree data D4 of 1st Embodiment. It is a figure which shows an example of the word tree data D5 of 1st Embodiment. It is a functional block diagram of search server 2 of a 2nd embodiment. It is a flowchart which shows an example of the flow of the search process of the search server 2 of 2nd Embodiment.

  Hereinafter, embodiments of an information processing device, an information processing method, and a program will be described with reference to the drawings.

<1. First Embodiment>
The first embodiment of the present invention will be described below. The information processing apparatus according to the present embodiment separates a query used for a search into words (hereinafter also referred to as “words”), and based on word tree data in which predefined words and categories are associated, The corresponding category is estimated, and information (search target) such as products belonging to the estimated category is output as a search result. The “category” in the present embodiment is information indicating a field to which content contents (products or services) provided as a search result for a query used for a search belong. In the present embodiment, an example in which the information processing apparatus is a search server that provides a search result of content such as products will be described.

<1-1. System configuration>
FIG. 1 is a functional block diagram of a search server 1 (information processing apparatus) according to this embodiment. The search server 1 is connected to the terminal device T and the like via the network NW. The network NW includes, for example, a WAN (Wide Area Network), a LAN (Local Area Network), the Internet, a dedicated line, a wireless base station, a provider, and the like.

  The terminal device T is operated by a user who uses a search service provided by the search server 1. The terminal device T is, for example, a computer device such as a personal computer, a mobile phone such as a smartphone, a tablet terminal, or a PDA (Personal Digital Assistant). In the terminal device T, a browser or an application program that operates based on a user operation transmits, for example, a query for requesting information provision to the search server 1 and receives information associated with the query from the search server 1.

  The search server 1 transmits the page information of the search result associated with the query input from the terminal device T to the terminal device T. The search server 1 generates search result page information using search data in which a query is associated with reference information for referring to a page of a site associated with the query. The reference information includes, for example, a URL (Uniform Resource Locator).

  The search server 1 includes, for example, a communication unit 10 (acquisition unit), a query separation unit 12 (separation unit), a determination unit 14, a first category estimation unit 16 (estimation unit), and a search result generation unit 18 (generation unit). ) And a storage unit 20. The communication unit 10 communicates with the terminal device T and the like via the network NW. The communication unit 10 receives a query from the terminal device T via the network NW, and transmits page information of the search result to the terminal device T. The communication unit 10 includes a communication interface such as a NIC, for example.

  The storage unit 20 includes, for example, search data D1 including reference information for referring to a page of a site associated with a query, category tree data D2 representing a relationship between categories in a tree structure, and a relationship between queries. Word tree data D3 and the like that are represented by a tree structure and include information on categories corresponding to each query are stored. The storage unit 20 is realized by, for example, a random access memory (RAM), a read only memory (ROM), a hard disk drive (HDD), a flash memory, or a hybrid storage device in which a plurality of these are combined. Further, a part or all of the storage unit 20 may be an external device accessible by the search server 1, such as a NAS or an external storage server.

  The query separation unit 12 analyzes the character string of the query acquired from the terminal device T and separates it into words. The query separation unit 12 separates the query into words, for example, using spaces, commas, and the like included in the query as delimiters.

  The determination unit 14 determines whether the word separated by the query separation unit 12 is defined in the word tree data D3.

  The first category estimation unit 16 sets the category corresponding to the word separated by the query separation unit 12 based on the word tree data D3 read from the storage unit 20 as the category corresponding to the query acquired from the terminal device T. presume. The first category estimation unit 16 applies the query acquired from the terminal device T based on the result of tracing the query acquired from the terminal device T in order from the query located at the top of the word tree data D3. Estimate the corresponding category. Details of the category estimation processing by the first category estimation unit 16 will be described later.

  The search result generation unit 18 performs a narrow search using the category estimated by the first category estimation unit 16 and generates a search result page that is a search result. For example, the search result generation unit 18 performs a narrow search using the estimated category for the search data D1 stored in the storage unit 20, and displays a search result page including information on products and the like belonging to the estimated category. Generate. The search result generation unit 18 transmits the generated search result page to the terminal device T via the communication unit 10.

  Each functional unit of the search server 1 is realized, for example, when a hardware processor such as a CPU (Central Processing Unit) executes a program (software) in a computer. Some or all of these components are hardware (circuits) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and GPU (Graphics Processing Unit). Part (including circuit)), or may be realized by cooperation of software and hardware.

<1-2. Word Tree Data>
Hereinafter, the word tree data D3 will be described. FIG. 2 is a diagram illustrating an example of the category tree data D2, and FIG. 3 is a diagram illustrating an example of the word tree data D3 defined based on the category tree data D2. The word tree data D3 is defined by a person such as a designer of the search server 1, for example.

  In the category tree data D2 illustrated in FIG. 2, “home appliance (A)” is defined as the category of the first hierarchy L1, and the category of the second hierarchy L2 positioned below the first hierarchy L1 “home appliance (A)”. “Air-conditioning appliance air-conditioning home appliance (A: A)” is defined as “the air-conditioner (A: A)” as the category of the third hierarchy L3 that is positioned below the second hierarchy L2 : A) ”is defined, and“ Air Conditioner (Home) ”(A: A: A: A) is defined as a category of the fourth hierarchy L4 that is positioned below the third hierarchy L3“ Air Conditioner (A: A: A) ” “Air conditioner for windows (A: A: A: B)”, “Air conditioner filter (A: A: A: C)”, “Remote control for air conditioner (A: A: A: D)”, and “Other air conditioner” Parts accessories (A: A: A: E) "are defined That. A symbol in parentheses indicates a category ID for identifying a category.

  In the category tree data D2, each category defined in the fourth hierarchy L4 indicates whether the category is associated with a classification of “main body”, “quasi-main body”, or “peripheral device”. The main body flag F is set. For example, the main body flag F “main body” is associated with “air conditioner (for home use) (A: A: A: A)”.

  The word tree data D3 is defined based on the category tree data D2. For example, the word tree data D3 is data (category dictionary data) that associates a query that is assumed to be used for a search (a query that is frequently used) and a category defined in the category tree data D2. is there. In the word tree data D3 illustrated in FIG. 3, the category ID “A: A: A” is associated with the query “air conditioner” as the first hierarchy H1. This category ID “A: A: A” corresponds to “air conditioner (A: A: A)” in the third hierarchy L3 in the category tree data D2 shown in FIG.

  In the word tree data D3, the category ID “A: A: A: A” is associated with the query “main body” of the second hierarchy H2 located below the first hierarchy H1 “air conditioner”. , A category ID “A: A: A: B” is associated with the query “for windows”, and a category ID “A: A: A: C” is associated with the query “filter”. A category ID “A: A: A: D” is associated with the query “remote control”, and a category ID “A: A: A: E” is associated with the query “component”.

  In the word tree data D3, in addition to the above query, the category “A: A: A: B” is associated with the query “air conditioner for windows” as the first hierarchy H1, The category ID “A: A: A: C” is associated with the query “air conditioner filter” as the first hierarchy H1. Further, the category ID “X: X” is associated with the query “automobile” as the first hierarchy H1, and the second hierarchy located below the first hierarchy H1 “automobile” is assumed to include similar queries. The category ID “X: X: X: Y” is associated with the query “air conditioner filter” of H2.

<1-3. Search process>
Hereinafter, the search process of the search server 1 will be described. FIG. 4 is a diagram illustrating an example of the flow of search processing of the search server 1. FIG. 5 is a diagram for explaining an outline of an example of the search process of the search server 1.

  First, the query separation unit 12 acquires a query input from the terminal device T via the communication unit 10 (S101). Next, the query separation unit 12 analyzes the query acquired from the terminal device T and separates it into words (S103). The query separation unit 12 separates the query into words, for example, using spaces, commas, and the like included in the query as delimiters.

  In the example illustrated in FIG. 5, the query separation unit 12 acquires a query “air conditioner for windows” that is input by the user via the terminal device T on the search page. ”And the word 2“ air conditioner ”.

  Next, the determination unit 14 determines whether or not the word separated by the query separation unit 12 is defined in the first hierarchy H1 (the highest layer) of the word tree data D3 (S105). In the example shown in FIG. 5, the word 1 “for window” does not exist in the first hierarchy H1 of the word tree data D3 shown in FIG. 3, but the word 2 “air conditioner” exists in the first hierarchy H1 of the word tree data. Exists. Further, the word 1 “for window” exists as the second hierarchy H2 positioned below the word 2 “air conditioner”. In this case, the determination unit 14 determines that the word separated by the query separation unit 12 is defined in the first hierarchy H1 of the word tree data D3.

  On the other hand, when all of the words separated by the query separation unit 12 do not exist in the first hierarchy H1 of the word tree data D3, the determination unit 14 determines that the word separated by the query separation unit 12 is the first word of the word tree data D3. It is determined that it is not defined in one hierarchy H1.

  When the determination unit 14 determines that the word separated by the query separation unit 12 is defined in the first hierarchy H1 of the word tree data D3, the first category estimation unit 16 uses the word tree data D3, A category corresponding to the word separated by the query separation unit 12 is estimated as a category corresponding to the query acquired from the terminal device T (S107). In the example illustrated in FIG. 5, the “window air conditioner” having the category ID “A: A: A: B” associated with the word 1 “for window” in the second hierarchy H2 lower than the word 2 “air conditioner”. It is estimated that the user intended the category.

  Next, with respect to the query acquired from the terminal device T, the search result generation unit 18 performs a narrowing search with the category estimated by the first category estimation unit 16 (S109). The search result generation unit 18 deletes words determined to be defined in the word tree data D3 by the determination unit 14 from the words included in the query acquired from the terminal device T, and searches the remaining words for the search query. As a result, a search for products and the like belonging to the category estimated for the search data D1 is performed.

  In the example illustrated in FIG. 5, it is determined that both the word 1 “for window” and the word 2 “air conditioner” included in the query acquired from the terminal device T are defined in the word tree data D3. In this case, since the search result generation unit 18 deletes the word 1 and the word 2, there is no search query for the search data D1. In this case, the search result generation unit 18 specifies only the category ID “A: A: A: B (window air conditioner)” estimated as the category intended by the user without specifying the search query, and the search data Search for D1.

  Next, the search result generation unit 18 generates a search result page based on the search result using the estimated category, and transmits it to the terminal device T (S111). When the determination unit 14 determines that the word separated by the query separation unit 12 is not defined in the word tree data D3, the search result generation unit 18 performs the terminal device without performing the category estimation process described above. A normal search process using the query acquired from T (a search process that does not narrow down the category) is performed to generate a search result page, which is transmitted to the terminal device T. Thus, the process of this flowchart ends.

  FIG. 6 is a diagram illustrating an example of a relationship between a query processed by the search server 1 and an estimated category. When the query acquired from the terminal device T is “air conditioner”, it is assumed that the user who inputs this query intends to search for “main body” of the air conditioner. For this reason, the “restrict body” flag is set in “air conditioner (A: A: A)” of the first hierarchy H1 in the word tree data D3 shown in FIG. This “main body narrowing down” flag is “the main body flag F” set in the category “air conditioner (for home use) (A: A: A: A)” in the fourth hierarchy L4 of the category tree data D2 shown in FIG. Associated with the "body". Therefore, when the query acquired from the terminal device T is “air conditioner”, the first category estimation unit 16 sets “air conditioner (for home use) (A: A: A: A)” as the category intended by the user. presume.

  Then, the search result generation unit 18 deletes the query “air conditioner” determined to be defined in the word tree data D3, and determines the category “air conditioner (for home use) (A: A) determined as the category intended by the user. : A: A) "is retrieved to generate a search result page.

  When the query acquired from the terminal device T is “air conditioner component remote control”, the query “air conditioner” exists as the first hierarchy H1 in the word tree data D3, and the second that is located below the query “air conditioner”. There are both queries “parts” and “remote control” as the hierarchy H2. In this case, the first category estimation unit 16 includes “other air conditioner component accessories” having the category ID “A: A: A: E” associated with the query “components” of the second hierarchy H2, and the second hierarchy H2. Both the category ID “A: A: A: D” and the “air conditioner remote control” associated with the query “remote control” are estimated as the categories intended by the user.

  Then, the search result generation unit 18 includes the query “air conditioner” determined to be defined as the first hierarchy H1, and the queries “parts” and “remote control” determined to be defined as the second hierarchy H2. , And information on products belonging to the category “other air conditioner parts accessories (A: A: A: E)” or the category “air conditioner remote control (A: A: A: D)” determined as the category intended by the user Is searched (OR search) to generate a search result page.

  When the query acquired from the terminal device T is “window air conditioner A company”, the query “air conditioner” exists as the first hierarchy H1 in the word tree data D3, and is located under the query “air conditioner”. The query “for windows” exists as the second hierarchy H2. In this case, the first category estimation unit 16 sets the category intended by the user as “window air conditioner” having the category ID “A: A: A: B” associated with the query “window” in the second hierarchy H2. presume.

  Then, the search result generation unit 18 deletes the word query “air conditioner” determined to be defined as the first hierarchy H1 and the query “window” determined to be defined as the second hierarchy H2. Then, by using the remaining query “Company A” as a search query, information on products belonging to the “window air conditioner” category is searched to generate a search result page.

  When the query acquired from the terminal device T is “filter”, this query does not exist in the first hierarchy H1 of the word tree data D3. In this case, the search result generation unit 18 performs a normal search process using the query “filter” without generating the category estimation process described above, and generates a search result page.

  When the query acquired from the terminal device T is “automobile air conditioner filter deodorization”, the query includes “air conditioner filter (A: A: A: C) present in the first hierarchy H1 of the word tree data D3. And “automobile (X: X)”. Furthermore, the query “air conditioner filter (X: X: X: Y)” exists as the second hierarchy H2 positioned below the query “automobile”. In this case, the first category estimation unit 16 estimates both “air conditioner filter (A: A: A: C)” and “air conditioner filter (X: X: X: Y)” as categories intended by the user. .

  In this case, it is determined which of the two estimated categories “air conditioner filter (A: A: A: C)” and “air conditioner filter (X: X: X: Y)” should be prioritized. Not stick. Therefore, the search result generation unit 18 does not delete the query that is determined to be defined in the word tree data D3, and uses “automobile air conditioner filter deodorization” as a search query. : C) "or" air conditioner filter (X: X: X: Y) "is searched (OR search), and a search result page is generated. That is, when the query acquired from the terminal device T includes two or more queries defined in the first hierarchy H1 of the word tree data D3, the search result generation unit 18 selects the word tree. The search processing for the search data D1 is performed using the query acquired from the terminal device T without deleting the query determined to be defined in the data D3.

  In addition, when the query acquired from the terminal device T is “down jacket” under the condition where the category tree data D4 as shown in FIG. 7 is defined, the query “down jacket” It is not possible to determine whether the jacket or ladies' down jacket is intended. In this case, as in the word tree data D5 shown in FIG. 8, the query “down jacket” is defined as the first hierarchy H1, and “men” and the query of the second hierarchy H2 positioned below the “down jacket” are defined as Define “Ladies”. Thereby, when the query acquired from the terminal device T is “down jacket”, the search result generation unit 18 selects the men's “down jacket (B: A: A)” or ladies' based on the word tree data D5. The product belonging to “down jacket (B: B: A)” can be searched (OR search).

  According to the search server 1 of this embodiment demonstrated above, the category corresponding to a query can be estimated appropriately and search accuracy can be improved further. In addition, by deleting a word that is determined to be defined in the word tree data and performing a refined search using the estimated category, the provider of the product etc., the information on its product etc. Therefore, it is not necessary to add a word that is assumed to be used for a search to the name of a product or the like for the purpose of being displayed at the top. In addition, by making it possible to narrow down the categories, information such as products in unintended categories will not be displayed when products are displayed in order from the lowest price, etc., improving the convenience of search sites be able to.

  In the above, examples of shopping sites in the home appliance field and fashion field have been described. Embodiments can be applied. Moreover, although the example which has one category corresponding to each query in the word tree data has been described, a plurality of categories may be associated with one query.

<2. Second Embodiment>
Hereinafter, a second embodiment of the present invention will be described. Compared with the first embodiment, the search server according to the present embodiment includes a second category estimation unit that estimates a category based on machine learning in addition to a first category estimation unit that estimates a category based on word tree data. Is different. For this reason, about the structure etc., the figure and related description which were demonstrated in 1st Embodiment are used, and detailed description is abbreviate | omitted.

<2-1. System configuration>
FIG. 9 is a functional block diagram of the search server 2 (information processing apparatus) of this embodiment. The search server 2 further includes a second category estimation unit 22 (second estimation unit) as compared with the search server 1 of the first embodiment. The first category estimation unit 16 (first estimation unit) performs a category estimation process based on a word tree (category dictionary data) created for a frequently used query, and therefore, for a query defined in the word tree. The estimation accuracy is high, and it is possible to cope with a query in which a plurality of categories are specified. However, the first category estimation unit 16 cannot be used for a query that is not defined in the word tree (that is, a query that is assumed to have a medium or low usage frequency, such as a product name). Therefore, in the present embodiment, the second category estimation unit 22 performs a category estimation process based on machine learning for a query that is not defined in the word tree. The second category estimation unit 22 includes, for example, a terminal category estimation unit 24 and an intermediate category estimation unit 26.

  The terminal category estimation unit 24 estimates the terminal (lowermost layer) category in the category tree. For example, the terminal category estimation unit 24 includes the query acquired from the terminal device T and the terminal category having the highest number of clicks among the categories clicked by the user in the search result page provided in response to the query (the user intended). The first model (hereinafter referred to as “terminal category model M1”) is generated by learning in advance the relationship with the category (correct category) that is considered to be stored in the storage unit 20. The terminal category estimation unit 24 estimates the terminal category corresponding to the query acquired from the terminal device T based on the terminal category model M1. The estimation processing by the terminal category estimation unit 24 has a sufficiently large number of queries processed from the terminal device T (or the number of clicks in the search result page provided according to the query acquired from the terminal device T). In this case, it is assumed that the estimation accuracy is high.

  The intermediate category estimation unit 26 estimates a category of an intermediate layer that can include a plurality of lower categories. For example, the intermediate category estimation unit 26 selects the intermediate category (the user's intention) among the queries acquired from the terminal device T and the category clicked by the user on the search result page provided in response to the query. The second model (hereinafter referred to as “intermediate category model M2”) is generated by learning in advance the relationship with the category (correct answer category) that is considered to have been stored, and stored in the storage unit 20. The intermediate category estimation unit 26 estimates an intermediate category corresponding to the query acquired from the terminal device T based on the intermediate category model M2.

  The above-described category estimation processing by the terminal category estimation unit 24 may reduce the estimation accuracy for a query targeting a category that is divided into ladies and mens. As described above, when it is assumed that the accuracy is lowered when the terminal category estimation unit 24 described above is used, the intermediate category estimation unit 26 that can include a plurality of lower categories is used. . For example, when the query acquired from the terminal device T is “duffle coat”, the intermediate category estimation unit 26 estimates a category that is “men's” and a category that is “ladies”. Next, the search result generation unit 18 narrows down the estimated category “Men” or “Ladies” and uses the search query “Duffle coat” (OR search of “Men” and “Ladies”). I do.

  The terminal category estimation unit 24 and the intermediate category estimation unit 26 include a multilayer neural network (Deep Neural Network: DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and the like. The deep learning technology used may be used.

<2-2. Search process>
Hereinafter, the search process of the search server 2 will be described. FIG. 10 is a diagram illustrating an example of the flow of search processing of the search server 2.

  First, the query separation unit 12 acquires a query input from the terminal device T via the communication unit 10 (S201). Next, the query separation unit 12 analyzes the query acquired from the terminal device T and separates it into words (S203).

  Next, the determining unit 14 determines whether or not the word separated by the query separating unit 12 is defined in the word tree data D3 (S205).

  When the determination unit 14 determines that the word separated by the query separation unit 12 is defined in the word tree data D3, the first category estimation unit 16 performs the query separation unit 12 based on the word tree data D3. The category corresponding to the separated word is estimated as the category corresponding to the query acquired from the terminal device T (S207).

  On the other hand, if the determination unit 14 determines that the word separated by the query separation unit 12 is not defined in the word tree data D3, the determination unit 14 further processes the same query in the past with respect to the query acquired from the terminal device T. Is determined to be greater than or equal to a threshold value (S213). This threshold value (an example of accuracy information) is a reference indicating the reliability of the terminal category estimation process corresponding to the query acquired from the terminal device T. If the number of queries processed from the terminal device T is greater than or equal to the threshold value, it is assumed that the end category is sufficiently learned for this query, and the accuracy of the estimation process by the end category estimation unit 24 is high. On the other hand, if the number of queries processed from the terminal device T is less than the threshold value, it is assumed that end category learning for this query is insufficient and the accuracy of the estimation processing by the end category estimation unit 24 is low. The number of processes for each query in the past is included in the search log L stored in the storage unit 20, for example.

  When the determination unit 14 determines that the number of processed queries acquired from the terminal device T is equal to or greater than the threshold, the terminal category estimation unit 24 selects the terminal category corresponding to the acquired query based on the terminal category estimation model M1. Estimate (S215). On the other hand, when the determination unit 14 determines that the number of queries processed from the terminal device T is not equal to or greater than the threshold, the intermediate category estimation unit 26 determines the end corresponding to the acquired query based on the intermediate category estimation model M2. A category is estimated (S217).

  Next, the search result generation unit 18 performs a refined search with respect to the query acquired from the terminal device T by the category estimated by any of the first category estimation unit 16, the end category estimation unit 24, and the intermediate category estimation unit 26. This is performed (S209). When performing a narrow search using the category estimated by the first category estimation unit 16, the search result generation unit 18 determines that the word tree data D3 is defined among the queries included in the acquired query. The retrieved query is deleted, and information such as products belonging to the category estimated using the remaining query as a search query is searched.

  Next, the search result generation unit 18 generates a search result page based on the search result using the estimated category and transmits it to the terminal device T (S211). Thus, the process of this flowchart ends.

  According to the search server 2 of this embodiment demonstrated above, the category corresponding to a query can be estimated appropriately and search accuracy can be improved further. When the category cannot be estimated based on the word tree data, the search accuracy can be further improved by estimating the category based on machine learning. When a plurality of categories are estimated by the intermediate category estimation unit 26, it is possible to provide a search result for each category.

  According to the embodiment described above, an acquisition unit that acquires a query input by a user, and a query acquired by the acquisition unit includes a relationship between queries expressed in a tree structure, and each of the queries And a determination unit that determines whether or not the category to be searched exists in the associated category dictionary data, and the determination unit determines that the query acquired by the acquisition unit exists in the category dictionary data, A first estimation unit that estimates a category corresponding to the query acquired by the acquisition unit based on the category dictionary data, and a query that the determination unit has acquired by the acquisition unit does not exist in the category dictionary data. When the determination is made, the acquisition unit performs the search based on the model used for learning the relationship between the query used for the search and the category. A second estimation unit that estimates a category corresponding to the acquired query; and a generation unit that generates information for displaying a search result belonging to the category estimated by the first estimation unit or the second estimation unit. Thus, it is possible to accurately estimate the category and improve the search accuracy.

  As mentioned above, although the form for implementing this invention was demonstrated using embodiment, this invention is not limited to such embodiment at all, In the range which does not deviate from the summary of this invention, various deformation | transformation and substitution Can be added.

1, 2 ... Search server (information processing device)
10. Communication unit (acquisition unit)
12 ... Query separation part (separation part)
14 ... Determination unit 16 ... First category estimation unit (first estimation unit)
18 ... Search result generation unit (generation unit)
20 ... Storage unit 22 ... Second category estimation unit (second estimation unit)
24 ... Terminal category estimation unit 26 ... Intermediate category estimation unit

Claims (9)

An acquisition unit for acquiring a query input by a user;
Determination that determines whether or not the query acquired by the acquisition unit exists in category dictionary data in which relationships between queries are represented in a tree structure and each of the queries and a category to be searched are associated with each other And
When the determination unit determines that the query acquired by the acquisition unit exists in the category dictionary data, a first estimation that estimates a category corresponding to the query acquired by the acquisition unit based on the category dictionary data And
When the determination unit determines that the query acquired by the acquisition unit does not exist in the category dictionary data, the determination unit is acquired by the acquisition unit based on a model that learns the relationship between the query used for the search and the category. A second estimation unit for estimating a category corresponding to the query,
An information processing apparatus comprising: a generation unit that generates information for displaying search results belonging to a category estimated by the first estimation unit or the second estimation unit. The second estimation unit includes
An end category estimation unit that estimates an end category corresponding to the query acquired by the acquisition unit and located at the end of the tree structure;
An intermediate category estimation unit corresponding to the query acquired by the acquisition unit and estimating an intermediate category located in an intermediate layer of the tree structure,
The information processing apparatus according to claim 1. The determination unit determines which of the terminal category estimation unit and the intermediate category estimation unit to perform the estimation process based on accuracy information of the first model used for the estimation process of the terminal category estimation unit. ,
The information processing apparatus according to claim 2. The determination unit determines whether or not the number of queries processed by the acquisition unit is equal to or greater than a threshold,
If the determination unit determines that the number of queries processed by the acquisition unit is greater than or equal to a threshold, the terminal category estimation unit performs an estimation process,
When the determination unit determines that the number of queries processed by the acquisition unit is not equal to or greater than a threshold, the intermediate category estimation unit performs an estimation process.
The information processing apparatus according to claim 3. A separation unit that separates the query acquired by the acquisition unit into words;
The first estimating unit estimates a category corresponding to the word separated by the separating unit as a category corresponding to the query obtained by the obtaining unit based on the category dictionary data.
The information processing apparatus according to any one of claims 1 to 4. The generation unit, when the determination unit determines that the query acquired by the acquisition unit is present in the category dictionary data, the word existing in the category dictionary data among the words separated by the separation unit And generating information for displaying the search result based on the result of searching the storage device that stores the information to be searched using the remaining words.
The information processing apparatus according to claim 5. The category dictionary data is defined based on a category tree representing a relationship between the search target categories in a tree structure.
The information processing apparatus according to any one of claims 1 to 6. Computer
Get the query entered by the user,
It is determined whether the acquired query is present in category dictionary data in which a relationship between queries is represented by a tree structure and each of the queries and a category to be searched are associated with each other.
If it is determined that the acquired query is present in the category dictionary data, a category corresponding to the acquired query is estimated based on the category dictionary data;
When it is determined that the acquired query does not exist in the category dictionary data, a category corresponding to the acquired query is estimated based on a model that has learned the relationship between the query used for the search and the category. ,
An information processing method for generating information for displaying a search result belonging to a category determined based on the category dictionary data or a category determined based on the model. On the computer,
Get the query entered by the user,
Determining whether the acquired query is present in category dictionary data in which relationships between queries are represented in a tree structure and each of the queries and a category to be searched are associated;
When it is determined that the acquired query is present in the category dictionary data, based on the category dictionary data, a category corresponding to the acquired query is estimated,
When it is determined that the acquired query does not exist in the category dictionary data, a category corresponding to the acquired query is estimated based on a model that has learned the relationship between the query used for the search and the category. ,
A program for generating information for displaying a search result belonging to a category determined based on the category dictionary data or a category determined based on the model.

Images