JP7461524B2 - Information processing device, information processing method, and program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and a program, and particularly relates to a technique for providing search results in response to an input query.

A query is a search request that is submitted to a search engine to cause the search engine to output desired search results.
For example, an EC (Electronic Commerce) site built on the web implements a search engine, accepts input of a query for a user to search for an item that the user wishes to purchase, causes the search engine to search for items tagged with tags that match the search key keywords contained in the query, and presents the user with a list of items obtained as the search results.

Many search engines allow multi-word queries that include multiple keywords in one query. Multi-word queries make it easier for users to reflect the user's intent from multiple perspectives in the query, but on the other hand, outputting only search results that completely match all of the multiple search keywords will result in fewer results than the original output. The desired search results will be omitted from the output, making it impossible to obtain the desired search results.

For this reason, search engines that allow multi-word queries usually use partial match, which outputs search results that match either a key phrase or keyword that forms part of a single query. By using search, the search scope is expanded and more search results are output.
However, each of the plurality of key phrases or keywords included in one query does not necessarily equally represent the user's intention of what he or she actually wants to purchase. Therefore, in a partial match search, there is a risk that search results that deviate greatly from the user's intention will be output, and the search accuracy may deteriorate.

Patent Document 1 discloses an information processing system that classifies queries into phrases or keywords, and selects phrases or keywords that are determined to express a user's desire from the classified phrases or keywords.
Specifically, in the information processing system of Patent Document 1, a query type classification unit classifies a query into either a phrase, a keyword, or any combination of these data, and a query type determination unit classifies a query into a desired dictionary. A database is searched to obtain query types and query type scores corresponding to classified phrases, keywords, or combination data, and the obtained phrases, keywords, or combination data with high query type scores are selected.

A query type is a classification pattern according to the nature and request of a phrase or keyword, such as "guidance type" or "request type," and a query type score is a numerical score that represents the accuracy of an estimated user request, which is predefined according to the query type. The request dictionary database stores the corresponding query type and query type score for each phrase, keyword, or combination data.

JP 2017-41030 A

According to the technique disclosed in Patent Document 1, it is possible to select a key phrase or keyword that is highly likely to correspond to the user's desire from among the key phrases or keywords that constitute one query.
However, with the technique disclosed in Patent Document 1, the query type and query type score must be defined in advance in the dictionary for each of the key phrases and keyword candidates that can be included in the query.
This increases the burden of registration and maintenance related to predefinitions, and it is not possible to obtain query type scores for query components that are not predefined, resulting in a lack of flexibility and a decrease in search accuracy. There is a risk of it getting lost.

The present invention has been made to solve the above problems, and its purpose is to provide an information processing device, an information processing method, and a program that can accurately identify components that match the user's intention from among the components of a query, without predefining each of the components included in the query.

In order to solve the above problem, one aspect of the information processing device according to the present invention includes a first query acquisition unit that acquires a first query, a query subset generation unit that divides the first query acquired by the first query acquisition unit into a plurality of tokens and generates a plurality of query subsets each consisting of one or more divided tokens, a second query acquisition unit that acquires a second query that shares the query subset generated by the query subset generation unit with the first query, a similarity calculation unit that calculates a similarity of the second query to the first query by referring to a query model that is generated by connecting the first query or the second query to an attribute of the query with a link and assigning a weight to the link based on a user's behavior history, a score calculation unit that calculates a score of the query subset based on the similarity of the second query calculated by the similarity calculation unit, and a query subset identification unit that identifies a query subset having a higher score calculated by the score calculation unit among the plurality of query subsets.

The similarity calculation unit may calculate the similarity as the degree to which the search result set of the second query is similar to the search result set of the first query.

The similarity calculation unit calculates the weight given to the link connected to the attribute shared between the query model of the first query and the query model of the second query. , the similarity of the second query may be calculated.

The similarity calculation unit may calculate the similarity of the second query by comparing the weights respectively assigned to pairs of links connected to the attributes shared between the query model of the first query and the query model of the second query, and adding up the weights having the smallest values for the attributes.

The similarity calculation unit may calculate the similarity of the second query by comparing information associated with items on which the user previously performed an operation between the first query and the second query.

The similarity calculation unit may generate the query model by setting information associated with the item as the attribute and assigning the weight based on the frequency of the operation.

The score calculation unit may calculate the score of the query subset by calculating a geometric mean of the similarities of all the second queries.

The second query acquisition unit acquires the second query for each of the plurality of query subsets generated by the query subset generation unit, and the score calculation unit acquires the second query for each of the plurality of query subsets. The score may be calculated.

The query subset generation unit may generate the query subset from multiple non-consecutive tokens in the character string of the first query.

The information processing device may further include an output control unit that controls output so that search results corresponding to the query subset identified by the query subset identification unit are presented preferentially.

One aspect of the information processing method according to the present invention is an information processing method executed by an information processing device, which includes the steps of obtaining a first query, and dividing the obtained first query into a plurality of tokens. , generating a plurality of query subsets consisting of one or more divided tokens, and obtaining a second query that shares the generated query subset with the first query;
By referring to a query model generated by connecting the first query or the second query to an attribute of the query with a link and assigning a weight to the link based on the user's action history. , calculating a similarity of the second query to the first query; calculating a score of the query subset based on the similarity of the second query; and a plurality of the queries. and identifying, among the subsets, a query subset with a higher calculated score.

One aspect of the information processing program according to the present invention is an information processing program for causing a computer to execute information processing, and the program includes: a first query acquisition process for acquiring a first query; a query subset generation process that divides the first query obtained by the first query acquisition process into a plurality of tokens and generates a plurality of query subsets each consisting of one or more of the divided tokens; and the query subset generation process. a second query acquisition process of acquiring a second query that shares the query subset generated by the process with the first query; and a second query acquisition process that acquires a second query that shares the query subset generated by the process with the first query; is connected to the attribute of the query with a link, and by referring to a query model generated by assigning weight to the link, the degree of similarity of the second query to the first query is calculated. a similarity calculation process, a score calculation process of calculating a score of the query subset based on the similarity of the second query calculated by the similarity calculation process, and among the plurality of query subsets, This is for executing a process including a query subset specifying process of specifying a query subset with a higher score calculated by the score calculation unit.

According to the present invention, it is possible to identify with a high degree of accuracy, from among the components of a query, a component that more closely matches the user's intent, without predefining each of the components contained in the query.
The above-mentioned objects, aspects, and advantages of the present invention, as well as objects, aspects, and advantages of the present invention not described above, will be understood by those skilled in the art from the following detailed description of the invention by referring to the accompanying drawings and the description of the claims.

FIG. 1 is a block diagram showing an example of the functional configuration of a search system including a keyword identification device according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating an example of a schematic processing procedure of keyword identification processing executed by the keyword identification device according to the present embodiment. FIG. 3 is a simplified example of a query model that defines the relationship between a query and an attribute and its weight, which is generated by the query model generation unit of the keyword identification device according to the present embodiment and stored in the query model storage unit. FIG. FIG. 4 is a conceptual diagram illustrating an example of calculating the similarity for a certain query in FIG. FIG. 5 is a conceptual diagram illustrating an example of calculating the similarity for another query in FIG. FIG. 6 is a flowchart showing an example of a detailed processing procedure of the single evaluation processing executed by the single evaluation unit of the keyword identification device according to the present embodiment. FIG. 7 is a conceptual diagram illustrating an example of calculating the similarity of another query that shares a certain single with respect to the original query, and calculating the score of the single based on the calculated similarity. FIG. 8 is a block diagram showing an example of a hardware configuration of the keyword identification device according to this embodiment.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. Among the constituent elements disclosed below, those having the same functions are given the same reference numerals, and their explanations will be omitted. The embodiments disclosed below are examples of means for realizing the present invention, and should be modified or changed as appropriate depending on the configuration of the device to which the present invention is applied and various conditions. The present invention is not limited to this embodiment. Furthermore, not all combinations of features described in this embodiment are essential to the solution of the present invention.

The keyword identification device according to the present embodiment divides a query to be input into a search engine into individual tokens that make up the query, generates a single consisting of one or more tokens, and For each, generate a query that includes the single.
The keyword identification device according to the present embodiment further refers to the query model generated based on the user's behavior history, calculates the similarity of each generated query to the original query, and uses the calculated similarity to calculate the similarity of each generated query to the original query. By calculating the score of the single, the single with a higher score is identified as a keyword that more closely matches the user's intention.

Below, an example will be described in which this embodiment is applied to a search engine implemented in EC (Electronic Commerce), for example, and in a use case in which a user inputs a multi-word query to purchase an item, a keyword in the query that best matches the user's intent is selected based on the user's behavior history; however, this embodiment is not limited to this and can be applied to searches for any purpose.

<Functional configuration of the keyword identification device>
FIG. 1 is a block diagram showing an example of a functional configuration of a keyword identification device 1 according to the present embodiment.
The keyword identification device 1 shown in FIG. 1 includes a query input unit 11, a query division unit 12, a query model generation unit 13, a single evaluation unit 14, and an output unit 15.
The keyword identification device 1 may be communicably connected to a client device (not shown) configured as a PC (Personal Computer) or the like via a network. In this case, the keyword identification device 1 may be implemented in a server, and the client device may provide a user interface when the keyword identification device 1 executes input and output of information with the outside, and may also include some or all of the components 11 to 15 of the keyword identification device 1.

Referring to FIG. 1, a search system is configured with a keyword identification device 1, a user behavior history DB (database) 2, a search engine 3, and a query model storage unit 4, and an example is shown in which the keyword identification device 1 accesses the user behavior history DB 2, generates a query model, stores it in the query model storage unit 4, and supplies the processing result of the keyword identification process to the search engine 3, but this embodiment is not limited to this. For example, the keyword identification device 1 may be implemented inside the search engine 3, or the user's behavior history may be obtained directly from the search engine 3.

The query input unit 11 inputs a query, which is a search request for causing the search engine 3 to execute a search, and supplies it to the query division unit 12 . The query input unit 11 may receive a query in real time from a client device connected via a network, or may obtain a query stored in a storage device in advance.
The query input unit 11 also receives input of various parameters necessary for executing keyword identification processing in the keyword identification device 1 . The query input unit 11 may receive input of various parameters via a user interface of a client device that is communicably connected to the keyword identification device 1 .

The query dividing unit 12 divides the query supplied from the query input unit 11 into a plurality of tokens. To split a multi-word query containing multiple keywords into multiple tokens, you can detect delimiters (spaces, commas, semicolons, etc.) between keywords written as strings in the query, or You may perform morphological analysis and break it down into parts of speech.
In this embodiment, the query dividing unit 12 further generates a plurality of singles to be evaluated (score calculation) by the single evaluation unit 14, which will be described later, from the divided plurality of tokens, and , is supplied to the single evaluation section 14.

A token is the minimum semantic unit that constitutes a key phrase or keyword included in a query, and one word corresponds to one token.
A single is a key phrase or keyword used for partial match search of a query, and is composed of one or more tokens (words).

As a non-limiting example, assume that the query as a search request is "travel coffee mug." In this case, the query "travel coffee mug" is split into three tokens "travel", "coffee", and "mug". Each token consists of one word, which is the smallest unit.
From these three tokens “travel”, “coffee”, and “mug”, we can create “travel” as a single consisting of two tokens (cluster of tokens).
"travel", "coffee", and "mug" are each generated as a single consisting of one token.

In this embodiment, as understood from the single "travel mug", a single can be created not only by combining consecutive words in the query, but also by combining words that are not consecutive (adjacent) in the query string. generated.
Thus, a single is a keyword or keyphrase made up of one token or any combination of tokens. That is, a single is any subset of a set of tokens, which is any subset of a query (query subset). A multi-word query generates multiple singles, ie, subsets of multiple queries.
Hereinafter, keywords and key phrases will be collectively referred to as "keywords." That is, a keyword is defined as one or more words that serve as a search key. By generating singles, the search range can be expanded and more search results can be output.

The query model generating unit 13 generates a query model based on the user's behavior history, and outputs the query model to the query model storage unit 4. The query model storage unit 4 is stored in a HDD (Hard Disk
The database is configured with a non-volatile storage device such as a hard disk drive (HDD) or a solid state drive (SSD), and stores the query model generated by the query model generation unit.
A query model is a graph that models a semantic relationship between a query and an attribute, the graph being composed of a query, one or more attributes associated with the query, and links connecting the query and each attribute. A weight indicating the degree of relevance between the query and the attribute connected to both ends of the link is assigned to the link. A query model may be a graph that connects multiple queries via one or more attributes associated with both of the multiple queries.

In this embodiment, the attributes associated with the query include, for example, the item genre (category), vendor (brand), item name or item ID, price range, and condition (new, new, used, etc.) that characterize already purchased items. , used items, etc.), tags (sale items, non-sale items, etc.), but is not limited thereto, and may be any information that can be acquired in association with a query.

In this embodiment, the query model generation unit 13 generates a query model based on the user's behavior history stored in the user behavior history DB2, for example, operations performed by the user on items in response to an input query and the frequency of these operations.
Note that operations performed by a user on an item include, but are not limited to, for example, an operation of clicking an image or link of an item, an operation of tapping an image of an item to view it, an operation of adding a "favorite" tab, an operation of purchasing an item, etc., and include any operation that may occur in the process of purchasing an item. Hereinafter, these operations will be simply referred to as "operations."

Specifically, the query model generating unit 13 generates a query model graph in which the input query and the item purchased in response to the query are connected by links, and each link is weighted according to the number of purchases, etc., and stores the generated graph in the query model storage unit 4. By tracking the user's actions, such as operations, it is possible to link the attributes of the purchased item to the query originally executed for the purchase, and also to determine the weight to be assigned to the link. The query model generated in this manner models the user's intention associated with the purchase of an item.
The query model generation unit 13 may refer to the user behavior history DB 2 and generate a query model in advance from queries previously submitted to the search engine 3 and operations performed by users in response to the queries. A query model may be generated for each user individually, or may be generated for a group of users grouped by age group, sex, occupation, past purchase amount, etc. Details of this query model will be described later with reference to Figs. 4 to 6.

The single evaluation unit 14 refers to the query model stored in the query model storage unit 4 and evaluates each of the multiple singles supplied from the query division unit 12 .
Specifically, the single evaluation unit 14 evaluates multiple singles by calculating a score for each single with respect to the original query based on a query (hereinafter also referred to as an "original query") that is the subject of the keyword identification process input to the query input unit 11 and multiple singles generated from the original query.
The score calculated by the single evaluation unit 14 is an index showing the degree to which each single matches the user's intention in the original query. The details of this single evaluation process will be described later with reference to FIG.

The output unit 15 outputs one or more singles for which a high score has been calculated by the single evaluation unit 14 to an output device such as a display device, and also supplies them to the search engine 3. The output unit 15 may also output various input data and various processing results of the keyword identification process executed by the keyword identification device 1, the query model stored in the query model storage unit 13, etc., via an output device such as a display device, as appropriate.

<Processing procedure for keyword identification processing>
FIG. 2 is a flowchart illustrating an example of a procedure for keyword identification processing executed by the keyword identification device 1 according to the present embodiment.
Note that each step in FIG. 2 is realized by the CPU reading out and executing a program stored in a storage device such as an HDD of the keyword identification device 1. Further, at least a portion of the flowchart shown in FIG. 2 may be realized by hardware. In the case of implementation using hardware, for example, a dedicated circuit may be automatically generated on an FPGA (Field Programmable Gate Array) from a program for implementing each step by using a predetermined compiler. Further, a Gate Array circuit may be formed in the same manner as an FPGA and realized as hardware. Alternatively, it may be realized by an ASIC (Application Specific Integrated Circuit). The same applies to each step in FIG. 3, which will be described later.

In S1, the query input unit 11 of the keyword identification device 1 accepts input of a query, which is a search request for causing the search engine 3 to execute a search. The input query may be a multi-word query including multiple keywords, or may be a single-word query including a single keyword. The query input unit 11 may receive a query from a client device connected via a network, or may obtain a query stored in advance in a storage device.
In S2, the query division unit 12 of the keyword identification device 1 divides the query supplied from the query input unit 11 into a plurality of tokens. Each of the plurality of tokens is a single keyword.

In S3, the query splitting unit 12 of the keyword identification device 1 generates multiple singles from the multiple tokens split in S2. The singles generated in S2 from the query input in S1 are a subset of the query, each consisting of one or multiple tokens. Each single is evaluated by the single evaluation unit 14 and assigned a score.

In S4, the single evaluation unit 14 of the keyword identification device 1 evaluates each single by calculating a score for each of the plurality of singles generated in S3.
Specifically, the single evaluation unit 14 refers to the query model stored in the query model storage unit 4, and for each single, inputs the query in S1 based on the behavior history of the user who inputs the query in S1. Calculate the score for the given query. The single score calculated here is an indicator of how well the query input in S1 matches the user's intention in purchasing the item, that is, how significant it is in expressing the user's intention. becomes. Details of the single evaluation process executed by the single evaluation unit 14 will be described later with reference to FIG.

In S5, the output unit 15 of the keyword identification device 1 selects one or more singles for which a high score has been calculated from among the plurality of singles for which scores have been calculated in S4, as a keyword that most prominently expresses the user's intention. It can be determined that In S5, the output unit 15 may supply the search engine 3 with one or more singles to be output.
The search engine 3 may, for example, sort the search results based on the query so that the search results corresponding to singles supplied from the output unit 15 are presented to the user more preferentially. Alternatively, if there are a small number of search results using the query input in S1 as a search key, or if the search range is too narrow, the search engine 3 may use other queries including singles supplied from the output unit 15 as a search key. You may perform the search again. This improves search accuracy and makes it possible to present search results that more closely match the user's intentions.

<Generation of query model and calculation of similarity between queries>
Generation of a query model and calculation of similarity between queries will be described with reference to FIGS. 3 to 5. In addition, in FIGS. 3 to 5, in order to simplify the explanation, an example in which each query is a single word query consisting of a single keyword will be explained, but the same applies in the case of a multi-word query containing multiple keywords. can be processed.

3 is a conceptual diagram for explaining, in a simplified manner, an example of a query model that specifies the relationship between a query and an attribute and its weight. Referring to FIG. 3, the query "pants" is connected to the attributes "kids", "womens", and "mens" by links. A weight of 0.2 is assigned to the link between the query "pants" and the attribute "kids", a weight of 0.3 is assigned to the link between the query "pants" and the attribute "womens", and a weight of 0.5 is assigned to the link between the query "pants" and the attribute "mens".
3 shows that a user who inputted the query "pants" to a search engine has a behavior history such as operations on items in the category of the attribute "kids", items in the category of the attribute "women", and items in the category of the attribute "men". In other words, each attribute is obtained by categorizing search results for the input query, and the similarity between queries indicates the degree of similarity between the search result set of the original query and the search result set of another query.
The weighting of each link correlates with the frequency of the operation. The weighting may be set according to the type of operation, for example, so that a higher weighting is given to a purchase operation than to a click operation.

Referring to FIG. 3, the queries "dress", "jeans", and "shirts" are connected to the attributes "kids", "womens", and "mens", respectively, by links. These queries "dress," "jeans," and "shirts" are other queries that share the attributes "kids," "womens," and "mens" with the query "pants," respectively. In FIG. 3, the query "pants" is the original query input in S1 of FIG. Assume that the other query is read from .

A weight of 0.4 is assigned to the link between the query "dress" and the attribute "kids", and a weight of 0.6 is assigned to the link between the query "dress" and the attribute "womens". On the other hand, the query "dress" is not associated with the attribute "mens", so it does not have a link to the attribute "mens".
The link between the query "jeans" and the attribute "kids" has a weight of 0.3, the link between the query "jeans" and the attribute "womens" has a weight of 0.2,
A weight of 0.5 is assigned to the link between the attribute "mens" and the attribute "ns".
A weight of 0.3 is assigned to the link between the query "shirts" and the attribute "kids", a weight of 0.4 is assigned to the link between the query "shirts" and the attribute "womens", and a weight of 0.3 is assigned to the link between the query "shirts" and the attribute "mens".

In Figure 3, the query models of the other queries "dress", "jeans", and "shirts" share attributes with the query model of the original query "pants", so the query model of the original query "pants" Each is similar.
FIG. 4 is a conceptual diagram of the query model of the original query "pants" and the query model of another query "shirts" extracted from FIG. 3.

In this embodiment, the single evaluation unit 14 calculates the degree of similarity for attributes shared between the original query and other queries by calculating the weight given to the link from each query to the attribute. do. For example, the single evaluation unit 14 may calculate the similarity by adding the minimum value (pairwise minimum) of the weights of the paired links.
Referring to FIG. 4, the minimum weight of a pair of links to the attribute "womens" is 0.3, the minimum weight of a pair of links to the attribute "kids" is 0.2, and the minimum weight of a pair of links to the attribute "mens" is 0.3. The minimum value of the weight of the pair of links is 0.3. Therefore, the sum of these three minimum values, 0.8, is the similarity of the query "shirts" to the original query "pants".

On the other hand, FIG. 5 is a conceptual diagram in which a query model for the original query "pants" and a query model for another query "jeans" are extracted from FIG. 3.
Referring to FIG. 5, the minimum value of the weight of a pair of links to the attribute "womens" is 0.2, the minimum value of the weight of a pair of links to the attribute "kids" is 0.2, and the minimum value of the weight of a pair of links to the attribute "mens" is 0.2. The minimum value of the weight of the pair of links is 0.5. Therefore, the sum of these three minimum values, 0.9, is the similarity of the query "jeans" to the original query "pants".
In the examples shown in FIGS. 3 to 5, the query "jeans" can be evaluated to have a higher degree of similarity to the original query "pants" than the query "shirts".

<Detailed procedure of single evaluation process>
FIG. 6 is a flowchart showing an example of a detailed processing procedure of the single evaluation process executed by the single evaluation unit 14 of the keyword identification device 1 according to this embodiment.
In S41, the single evaluation unit 14 of the keyword identification device 1 determines the attributes and weights of the original query input in S1 of FIG.
Specifically, the single evaluation unit 14 refers to the query model storage unit 4, and if the query model of the original query input in S1 has already been stored in the query model storage unit 4, reads the query model of the original query from the query model storage unit 4, skips S42, and proceeds to S43.

On the other hand, if the query model of the original query is not stored in the query model storage unit 4, the single evaluation unit 14 may refer to the user behavior history DB 2 to acquire attributes associated with items previously purchased by the user who input the original query in S1, and determine these as the attributes of the original query, or may estimate the attributes of the item to be purchased from information on the page of the EC site displayed when the original query was input. A predetermined initial value may be set for each attribute as the weight. Alternatively, the single evaluation unit 14 may, for example, read out the query model of the query that shares the most tokens with the original query from the query model storage unit 13, and determine or infer the attributes and weight of the read out query model as the attributes and weight of the original query.
In addition, in the case of a new user with no past behavioral history, the initial query model of the original query may be inferred from the behavioral history and query models of other users or user groups in categories similar to the user's information that can be obtained when registering as a member.

In S42, the single evaluation unit 14 of the keyword identification device 1 causes the query model generation unit 13 to generate a query model by associating the attributes and weights determined in S41 with the original query input in S1. The query model generation unit 13 may store the query model generated in S42 in the query model storage unit 4.

In S43, for each single generated in S3 of FIG. 2, the single evaluation unit 14 of the keyword identification device 1 obtains, from the query model storage unit 4, other queries that share the single, i.e., all tokens contained in the single, with the original query.
The single evaluation unit 14 may obtain other queries that share the single with the original query and at least one attribute with the original query, because the similarity between queries that do not share an attribute is 0 according to the similarity calculation of the examples of FIG.

In S44, the single evaluation unit 14 of the keyword identification device 1 calculates the similarity of each of the other queries acquired in S43 to the original query.
Specifically, the single evaluation unit 14 calculates the similarity of the other query to the original query by calculating the weights assigned to the paired links connecting the two queries for an attribute shared by the query model of the original query and the query model of the other query. As already described with reference to Figures 4 and 5, the single evaluation unit 14 may calculate the similarity by, for example, adding the minimum value of the weights of the paired links of the attribute.

In S45, the single evaluation unit 14 of the keyword identification device 1 calculates the score of the single to be evaluated based on the similarity calculated in S44 for all other queries acquired in S43.
Specifically, the single evaluation unit 14 calculates the average of the plurality of similarities calculated in S44, for example, a geometric mean, for all of the plurality of queries acquired in S43, as the single to be evaluated. It may be calculated as a score of

In S46, the single evaluation unit 14 of the keyword identification device 1 determines whether or not all of the singles generated in S3 of FIG. 2 have been evaluated.
If there are unprocessed singles (S46: N), the process returns to S43 and repeats the processes from S43 to S46. On the other hand, if there are no unprocessed singles (S46: Y), the process proceeds to S5 in Fig. 2 and one or more singles with higher scores are output as keywords that match the user's intention.
In this way, the keyword identification device 1 of this embodiment can identify keywords without predefining tokens or singles by evaluating the user behavioral history associated with each of the other queries that contain these tokens, rather than evaluating the individual tokens themselves that make up the original query.

FIG. 7 is a conceptual diagram illustrating an example of calculating the similarity of another query that shares a certain single with respect to the original query, and calculating the score of the single based on the calculated similarity.
Referring to FIG. 7, it is assumed that the original query input in S1 of FIG. 2 is {“a”, “b”, “c”}. In this case, in S2 of FIG. 2, the original query is divided into three tokens {“a”}, {“b”}, {“c”}, and in S3 of FIG. 2, the original query is divided into six single tokens {“
a”, “b”}, {“a”, “c”}, {“b”, “c”}, {“a”}, {“b”}, and {“c”} are generated.

For each of these six singles, all other queries that include that single are obtained in S43 of Fig. 6. For example, for single {"a", "b"}, queries {"a", "b", "d"} and {"a", "b", "c"} are obtained. Query {"a", "b", "c"} is the same query as the original query.
For each of these other queries {"a", "b", "d"} and {"a", "b", "c"} obtained for the single {"a", "b"}, the similarity to the original query {"a", "b", "c"} is calculated in S44 of FIG. 6.

7, query {"a", "b", "d"} shares attributes α1 and α2 with original query {"a", "b", "c"}, and the minimum weighting value assigned to a pair of links connecting to attribute α1 is 0.2, and the minimum weighting value assigned to a pair of links connecting to attribute α2 is 0.3. Therefore, the similarity of query {"a", "b", "d"} to original query {"a", "b", "c"} is calculated as 0.5, which is the sum of 0.2 and 0.3.
On the other hand, query {"a", "b", "c"}, which is the same as the original query, shares all attributes α1, α2, and α3 with the original query {"a", "b", "c"}, and the minimum weights assigned to the paired links connecting each attribute are 0.3. Therefore, the similarity of query {"a", "b", "c"} to the original query {"a", "b", "c"} is calculated as 0.9.
Therefore, in S45 of FIG. 6, the score of the single {"a", "b"} is calculated as 0.67, which is the geometric mean of 0.5 and 0.9.
By repeating the above process for all of the singles generated in S3 of FIG. 2, one or more singles with the highest scores can be identified.

<Hardware configuration of keyword identification device>
FIG. 8 is a diagram showing a non-limiting example of the hardware configuration of the keyword identifying device 1 according to the present embodiment.
The keyword identification device 1 according to this embodiment can be implemented on any single or multiple computers, mobile devices, or any other processing platform.
Referring to FIG. 8, an example is shown in which the keyword identification device 1 is implemented in a single computer, but the keyword identification device 1 according to the present embodiment is implemented in a computer system including a plurality of computers. It's fine. A plurality of computers may be connected to each other through a wired or wireless network so that they can communicate with each other.

As shown in FIG. 8, the keyword identification device 1 may include a CPU 81, a ROM 82, a RAM 83, an HDD 84, an input section 85, a display section 86, a communication I/F 87, and a system bus 88. The keyword identification device 1 may also include an external memory.
A CPU (Central Processing Unit) 81 centrally controls operations in the keyword identification device 1, and controls each component (82 to 87) via a system bus 88, which is a data transmission path.

A ROM (Read Only Memory) 82 is a nonvolatile memory that stores control programs and the like necessary for the CPU 81 to execute processing. Note that the program may be stored in a nonvolatile memory such as an HDD (Hard Disk Drive) 84 or an SSD (Solid State Drive), or an external memory such as a removable storage medium (not shown).
A RAM (Random Access Memory) 83 is a volatile memory and functions as the main memory, work area, etc. of the CPU 81. That is, the CPU 81 loads necessary programs and the like from the ROM 82 into the RAM 83 when executing processing, and implements various functional operations by executing the programs and the like.

The HDD 84 stores, for example, various data and various information required when the CPU 81 performs processing using a program. The HDD 84 also stores, for example, various data and various information obtained when the CPU 81 performs processing using a program.
The input unit 85 is composed of a keyboard and a pointing device such as a mouse.
The display unit 86 is configured with a monitor such as a liquid crystal display (LCD). The display unit 86 may provide a GUI (Graphical User Interface) that is a user interface for inputting various parameters used in the keyword identification process, communication parameters used in communication with other devices, and the like, to the keyword identification device 1.

The communication I/F 87 is an interface that controls communication between the keyword identification device 1 and an external device.
The communication I/F 87 provides an interface with a network and executes communication with an external device via the network. Various data, various parameters, etc. are transmitted and received between the external device and the communication I/F 87. In this embodiment, the communication I/F 87 may execute communication via a wired LAN (Local Area Network) or a dedicated line that complies with a communication standard such as Ethernet (registered trademark). However, the network that can be used in this embodiment is not limited to this, and may be configured as a wireless network. This wireless network includes wireless PANs (Personal Area Networks) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). It also includes wireless LANs (Local Area Networks) such as Wi-Fi (Wireless Fidelity) (registered trademark) and wireless MANs (Metropolitan Area Networks) such as WiMAX (registered trademark). It also includes wireless WANs (Wide Area Networks) such as LTE/3G, 4G, and 5G. Note that the network only needs to be able to connect devices to each other and communicate with each other, and the communication standard, scale, and configuration are not limited to those described above.

At least some of the functions of each element of the keyword identification device 1 shown in FIG. 1 can be realized by the CPU 81 executing a program. However, at least some of the functions of each element of the keyword identification device 1 shown in FIG. 1 may be operated as dedicated hardware. In this case, the dedicated hardware operates under the control of the CPU 81.

As described above, according to this embodiment, the keyword identification device divides a query to be input to a search engine into individual tokens that constitute the query, generates shingles consisting of one or more tokens as subsets of the query, and for each of the generated shingles, generates a query including the shingle.
The keyword identification device of this embodiment further calculates the similarity of each generated query to the original query by referring to the query model generated based on the user's behavioral history, and calculates a score for the single from the calculated similarity, thereby identifying the single with a higher calculated score as the keyword that better matches the user's intention.

Therefore, without predefining each of the tokens and shingles contained in the query, it is possible to flexibly and easily identify keywords that are more representative of the user's intent from among the shingles (key phrases and keywords) that make up the query.
This allows desired search results to be presented in response to a query, contributing to improved search accuracy.

Note that although specific embodiments have been described above, these embodiments are merely illustrative and are not intended to limit the scope of the present invention. The apparatus and methods described herein may be embodied in forms other than those described above. In addition, appropriate omissions, substitutions, and changes may be made to the embodiments described above without departing from the scope of the present invention. Such omissions, substitutions, and modifications are included within the scope of the claims and their equivalents, and fall within the technical scope of the present invention.

1...Keyword identification device, 2...User behavior history DB, 3...Search engine, 4...Query model storage unit, 11...Query input unit, 12...Query division unit, 13...Query model generation unit, 14...Single evaluation unit, 15...Output unit, 81...CPU, 82...ROM, 83...RAM, 84...HDD, 85...Input unit, 86...Display unit, 87...Communication I/F, 88...Bus

Claims (12)

a first query acquisition unit that acquires a first query;
a query subset generation unit that divides the first query acquired by the first query acquisition unit into a plurality of tokens and generates a plurality of query subsets each consisting of one or more divided tokens;
a second query acquisition unit that acquires a second query that shares the query subset generated by the query subset generation unit with the first query;
By referring to a query model generated by connecting the first query or the second query to an attribute of the query with a link and assigning a weight to the link based on the user's action history. , a similarity calculation unit that calculates the similarity of the second query to the first query;
a score calculation unit that calculates a score of the query subset based on the similarity of the second query calculated by the similarity calculation unit;
An information processing device comprising: a query subset specifying unit that specifies, among the plurality of query subsets, a query subset with a higher score calculated by the score calculating unit. The information processing apparatus according to claim 1 , wherein the similarity calculation unit calculates the similarity as a degree to which the search result set of the second query is similar to the search result set of the first query. The similarity calculation unit calculates the weight given to the link connected to the attribute shared between the query model of the first query and the query model of the second query. , the information processing apparatus according to claim 1 or 2, wherein the similarity of the second query is calculated. 4. The information processing device according to claim 3, wherein the similarity calculation unit calculates the similarity of the second query by comparing the weights respectively assigned to pairs of the links connected to the attributes shared between the query model of the first query and the query model of the second query, and adding up the weights having smaller values for the attributes. The similarity calculation unit calculates the similarity of the second query by comparing information associated with an item on which the user has performed an operation in the past between the first query and the second query. The information processing device according to any one of claims 1 to 4, wherein the information processing device calculates . The information processing device according to claim 5 , wherein the similarity calculation unit generates the query model by setting information associated with the item as the attribute and assigning the weight based on the frequency of the operation. The information processing device according to claim 1 , wherein the score calculation unit calculates a geometric mean of the similarities of all the second queries to calculate the score of the query subset. The second query acquisition unit acquires the second query for each of the plurality of query subsets generated by the query subset generation unit,
The information processing device according to any one of claims 1 to 7, wherein the score calculation unit calculates the score for each of the plurality of query subsets. The information processing device according to claim 1 , wherein the query subset generation unit generates the query subset from a plurality of tokens that are not consecutive in a character string of the first query. Any one of claims 1 to 9, further comprising an output control unit that controls output so that search results corresponding to the query subset specified by the query subset specifying unit are preferentially presented. The information processing device described in . An information processing method executed by an information processing device,
obtaining a first query;
Dividing the acquired first query into a plurality of tokens and generating a plurality of query subsets each including one or more of the divided tokens;
obtaining a second query that shares the generated query subset with the first query;
calculating a similarity of the second query to the first query by referring to a query model generated by connecting the first query or the second query to an attribute of the query with a link and assigning a weight to the link based on a user's behavior history;
calculating a score for the subset of queries based on the similarity measure of the second query;
and identifying a query subset having a higher calculated score from among the plurality of query subsets. An information processing program for causing a computer to execute information processing, the program comprising:
a first query acquisition process for acquiring a first query;
a query subset generation process of dividing the first query acquired by the first query acquisition process into a plurality of tokens and generating a plurality of query subsets each including one or more of the divided tokens;
a second query acquisition process for acquiring a second query that shares the query subset generated by the query subset generation process with the first query;
a similarity calculation process for calculating a similarity of the second query to the first query by referring to a query model generated by connecting the first query or the second query to an attribute of the query with a link and assigning a weight to the link based on a user's behavior history;
a score calculation process of calculating a score of the query subset based on the similarity of the second query calculated by the similarity calculation process;
a query subset identification process for identifying a query subset having a higher score calculated by the score calculation process from among the plurality of query subsets.

Images