JP7088693B2 - Information processing equipment, information processing methods, and programs - Google Patents

Description

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

Conventionally, a crawler that collects data (documents, images, etc.) from the Web and automatically creates a database of the collected data is known (see Patent Document 1). Crawlers collect data from web pages with various IP addresses by following links in web pages. The data collected by the crawler is stored in the web information database.

On the other hand, when the search engine receives the search word input by the user, the information of the web page or application page related to the received search word (for example, URL: Uniform Resource Locator) is acquired from the web information database and acquired. Output a list of information as a search result.

Japanese Unexamined Patent Publication No. 2012-69171

Matthew D. Hoffman et al., “Online Learning for Latent Dirichlet Allocation”, [online], [Searched February 27, 2018], Internet <URL: https://papers.nips.cc/paper/3902-online-learning -for-latent-dirichlet-allocation.pdf ＞

When the search engine is operated to output the data related to the search word (hereinafter, also referred to as "query" or "search query") entered by the user together with the search result in order to improve the user satisfaction. There is. In order to realize this, a database in which text data and an entity are stored in association with each other may be used. In the database, when text data associated only with a specific entity is input as a query, that specific entity is naturally output together with the search result. On the other hand, when one text data is associated with a plurality of entities, information about an inappropriate entity not intended by the user may be output together with the search result. Also, if the query entered in one search request processing contains multiple words, it is not possible to determine which word represents the entity, and information about the inappropriate entity is output along with the search results. May be done.

The present invention has been made in consideration of such circumstances, and one of the objects of the present invention is to provide an information processing apparatus capable of suppressing output of inappropriate information together with a search result.

One aspect of the present invention includes each of two or more words input as a search query, a first index value calculation unit that calculates a first index value indicating the degree of association with an entity, and the word as a context. The second index value calculation unit that calculates the second index value indicating the degree of association between the entity and each of the words, the first index value calculated by the first index value calculation unit, and the second index. It is an information processing apparatus including a determination unit that determines an entity that is highly related to the word based on the second index value calculated by the value calculation unit.

According to one aspect of the present invention, it is possible to suppress the output of inappropriate information together with the search result.

It is a figure which shows the use environment and the configuration of the knowledge data server which concerns on this embodiment. It is a figure which shows an example of the knowledge graph which concerns on this embodiment. It is a figure which shows an example of the web page of "baseball player A" which concerns on this embodiment. It is a figure which shows an example of the web page of "baseball team B" which concerns on this embodiment. It is a figure which shows an example of the content information which concerns on this embodiment. It is a figure which shows an example of the search result web page which concerns on this embodiment. It is a figure which shows an example of the related web page collected by the collection part which concerns on this embodiment. It is a figure which shows another example of the related web page collected by the collection part which concerns on this embodiment. It is a figure which shows another example of the related web page collected by the collection part which concerns on this embodiment. It is a figure which shows another example of the web page of "baseball player A" which concerns on this embodiment. It is a figure which shows an example of the content information and the 1st probability which concerns on this embodiment. It is a figure explaining an example of the context collection process by the context collection part which concerns on this embodiment. It is a figure explaining another example of the context collection process by the context collection part which concerns on this embodiment. It is a figure which shows an example of the context information which concerns on this embodiment. It is a figure explaining an example of the calculation processing of the frequency information of a context by the 2nd probability calculation part which concerns on this embodiment. It is a flow chart which shows an example of the online operation of the knowledge data server of this embodiment. It is a figure explaining an example of the online operation of the knowledge data server of this embodiment. It is a flow chart which shows an example of the operation which calculates the 1st probability of the knowledge data server of this embodiment. It is a flow chart which shows an example of the operation which calculates the 2nd probability of the knowledge data server of this embodiment. It is a flow chart which shows an example of the detailed operation which calculates the 2nd probability of the knowledge data server of this embodiment.

Hereinafter, embodiments of the information processing apparatus, information processing method, and program of the present invention will be described with reference to the drawings. In the present embodiment, the information processing apparatus will be described as forming a part of the knowledge data server. The knowledge data server is, for example, a server that collects data to be collected (for example, image or text data) and generates a knowledge graph based on the collected data. In the present embodiment, the knowledge graph is data in which an entity, a class, and a property are associated with the location information of a web page in which related data related to the entity is described. An entity may represent, for example, an entity of an object (eg, an object that exists in the real world), or may be defined in the concept of an object (eg, in the real world or virtual world). It may represent a concept). For example, when the object is the concept of "building", the entity may represent an entity such as "○○ tower" or "○○ building". Further, for example, when the object is the concept of "economics", the entity may represent an insubstantial abstract concept such as "microeconomics" or "macroeconomics". Hereinafter, embodiments will be described in detail.

<Embodiment>
<1-1. Knowledge data server usage environment>
FIG. 1 is a diagram showing a usage environment and configuration of the knowledge data server 100 according to the present embodiment. The knowledge data server 100 is communicably connected to the terminal device 200, the web server 300, the crawl target device DV, and the network NW. The network NW means the World Wide Web, and is a system using HTML documents and the like that are standardly used on the Internet and intranets. The network NW may further include a radio base station, a provider device, a dedicated line, and the like.

The terminal device 200 is a device used by a user, and is, for example, a mobile phone such as a smartphone, a tablet computer, a notebook computer, a desktop computer, or the like. The web server 300 is a server that uses a search engine to generate a web page for displaying search results and provides it to the terminal device 200.

The knowledge data server 100 includes, for example, a control unit 110 and a storage unit 120. The storage unit 120 is realized by, for example, a RAM (Random Access Memory), an HDD (Hard Disk Drive), 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 120 may be an external device such as NAS (Network Attached Storage) or an external storage server that can be accessed by the knowledge data server 100. The storage unit 120 stores, for example, a knowledge graph (hereinafter, knowledge graph D1), content information D2, and context information D3.

FIG. 2 is a diagram showing an example of the knowledge graph D1 according to the present embodiment. The things described in the knowledge graph D1 are defined by the ontology. An ontology is a definition of a class and a property of an object, and is a collection of constraints that hold between the class and the property.

A class is a group of things that have the same properties in an ontology. What the nature of an object is, that is, which class the object belongs to, is determined by the properties described below.

For example, things that have a beak, an oviparous vertebrate, and forelimbs that are wings are classified in the "bird" class. Also, in the class of "birds", things that have the property of not being able to fly are classified into lower classes such as "penguins" and "ostriches". In this way, the class system may have a hierarchical structure having a higher-lower relationship. The nature of the upper class is inherited by the lower class. In the example above, the "bird" class's "beak-bearing, oviparous vertebrate with winged forelimbs" property is a lower class property of "penguins" and "ostriches". Will also be included. For example, if the class name for identifying the class is a class "bird", the class name may be represented by the character string "bird". The class name does not necessarily have to represent a meaning, and for example, even in the class "bird", a character string indicating mere identification information such as "information 1" or "C1" may be assigned. .. The above-mentioned entity, that is, an entity, shall belong to any of the classes contained in the class system defined by the ontology.

Properties are attributes that describe the nature and characteristics of things and the relationships between classes. For example, a property may be an attribute that has the property of "having ... as a component of the body" or the property of "living in", or "a class is a higher class and a certain class is a lower class". It may be an attribute indicating the relationship between the upper and lower levels of "class". As for the property name for identifying the property, the property name itself may or may not represent the meaning, as in the class name described above.

The knowledge graph D1 is a directed graph in which the above-mentioned class is represented as a node and the above-mentioned properties are represented as labeled and directional edges. With such a graph structure, information about things can be discriminated by nodes, and relationships between things can be discriminated by edges.

In the knowledge graph D1 of the example shown in FIG. 2, the entity E1 and the entity E2 belong to a class having a property name of "affiliated team". In the present embodiment, each entity is associated with information that can identify each entity (hereinafter, entity identification information EID), an entity name, and location information of a web page of the entity. The location information is information for specifying a position on the Web, and is, for example, a URL. The web page indicated by the location information contains the relevant data of the entity. In the knowledge graph D1 shown in FIG. 2, the entity E1 is the location of a web page in which the entity identification information EID "0001", the entity name "baseball player A", and related data related to the entity are described. It is associated with the URL "http: // encyclopedia web page / baseball player A" which is information. Further, the entity E2 is the location information of the web page in which the entity identification information EID "0002", the entity name "baseball team B", and the related data related to the entity are described, "http: //". It is associated with the URL "Encyclopedia Web Page / Baseball Team B". In the following description, the related data related to the entity will be described as the related data of the entity. In addition, the web page in which the data related to the entity is described is also described as the web page of the entity.

FIG. 3 is a diagram showing an example of a web page of “baseball player A” according to the present embodiment. The web page of "baseball player A" contains data related to "baseball player A". Further, FIG. 4 is a diagram showing an example of a web page of “baseball team B” according to the present embodiment. The web page of "Baseball Team B" contains data related to "Baseball Team B".

FIG. 5 is a diagram showing an example of the content information D2 according to the present embodiment. The content information D2 is information in which an entity and text data are associated with each other. The content information D2 is generated based on the collection result of collecting the related data contained in the web page or the search log of the search engine. The content information D2 is, for example, information in which text data included in a web page of an entity is collected and the collected text data is associated with the entity. Further, the content information D2 is, for example, information in which when an entity is input as a query, the text data input together with the entity is associated with the entity. In the example content information D2 shown in FIG. 5, the entity "baseball player A" is associated with the text data "3000 hits", and the entity "baseball team B" is referred to as "2017 championship". Text data is associated.

<1-2. Search result web page ＞
FIG. 6 is a diagram showing an example of a search result web page according to the present embodiment. As shown in FIG. 6, the display unit 210 of the terminal device 200 displays a search result web page including a query input area 211, a knowledge panel 220, and a search result 230. The knowledge panel 220 includes the entity 221 corresponding to the query, the image 222 associated with the entity 221 and the relevant information 223 associated with the entity 221. The user uses the terminal device 200 to input a query on the search page displayed on the display unit 210 of the terminal device 200. A query is a search word or a combination of a plurality of search words. The terminal device 200 sends the query entered by the user to the web server 300.

The web server 300 transmits the query received from the terminal device 200 to the knowledge data server 100. The knowledge data server 100 sends information related to the entity corresponding to the received query to the web server 300. For example, when the received query is included in the entity of the content information D2, the knowledge data server 100 determines that the entity is the entity associated with the received query. The knowledge data server 100 determines an entity associated with a received query (text data), for example, based on the content information D2. Further, the knowledge data server 100 transmits the location information associated with the determined entity to the web server 300 based on the knowledge graph D1. Based on the received location information, the web server 300 extracts the relevant data of the entity from the web pages indicated by the location information. The web server 300 generates a search result web page (eg, FIG. 6) that includes the relevant data of the extracted entity (in this example, the information shown in the knowledge panel 220).

In the example shown in FIG. 6, "3000 hits" is input as a query, and an image and various information related to "3000 hits" are displayed on the knowledge panel 220. On the knowledge panel 220, for example, an image 222 of "baseball player A" who achieved "3000 hits" and related information 223 such as the date of birth and birthplace of "baseball player A" are displayed. In this way, by displaying not only the search result 230 but also the knowledge panel 220 on the display unit 210, the user satisfaction with the search result can be improved.

<1-3. Configuration of Knowledge Data Server 100>
Returning to FIG. 1, the control unit 110 is realized by, for example, a hardware processor such as a CPU (Central Processing Unit) executing a program (software). The control unit 110 is, for example, a selection unit 111, a related page collection unit 112, a first probability calculation unit 113 (first index value calculation unit), a context collection unit 114, and a second probability calculation unit 115 (second probability calculation unit). It includes an index value calculation unit, a frequency information calculation unit), a location information acquisition unit 116 (determination unit), and a communication I / F (Interface) 117. In addition, some or all of these components (excluding the contained storage unit) are LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), GPU (Graphics Processing). It may be realized by hardware (circuit unit; including circuitry) such as Unit), or it may be realized by the cooperation of software and hardware. The communication I / F 117 mediates communication between each functional unit via the network NW and other devices (for example, a crawl target device (hereinafter, crawl target device DV), a terminal device 200, and a web server 300). ..

The selection unit 111 selects arbitrary text data (hereinafter referred to as target text data) for which the probability is to be calculated from the content information D2. The related page collecting unit 112 collects a web page (hereinafter referred to as a related web page) related to the target text data selected by the selection unit 111 via the network NW. The web page collected by the related page collection unit 112 is, for example, an encyclopedia web page that can be edited by a user, or a search result web page when the target text data is used as a query. Hereinafter, the encyclopedia web page that can be edited by the user is referred to as an encyclopedia web page.

FIG. 7 is a diagram showing an example of an encyclopedia web page collected by the related page collection unit 112. In the example shown in FIG. 7, the related page collection unit 112 describes the target text data of "3000 hits" associated with the entity of "baseball player A" indicated by the content information D2 in the encyclopedia web page. Collect related web pages. The related page collection unit 112 collects collection target data including HTML (HyperText Markup Language) data of an encyclopedia web page from a crawl target device DV via a network NW. The related page collecting unit 112 is realized by a so-called crawler program. The data to be collected is data that is on the network NW (in the storage area of the device DV to be crawled) and can be viewed by a browser. The data to be collected is not limited to the browser, and may be data reproduced by the application program. The data to be collected is, for example, a related web page containing textual data indicating the source of HTML. The related page collection unit 112 collects all the related web pages in which the target text data is described among the encyclopedia web pages, if any.

Further, FIG. 8 is a diagram showing an example of a search result web page collected by the related page collection unit 112. In the example shown in FIG. 8, in the search result web page, the query "3000 hits" (text TX21 shown) in which the entity "baseball player A" indicated by the content information D2 is displayed as the search result is input. The search result is shown. The query is an example of target text data.

Returning to FIG. 1, the first probability calculation unit 113 performs the probability calculation process for the set of the target text data and the entity based on the related web page collected by the related page collection unit 112. The first probability calculation unit 113 calculates a probability (P (entity | name)) (hereinafter, first probability) indicating the degree of association of each entity when the target text data (name of the entity) is given. .. The related page collecting unit 112 calculates, for example, the value of P (e | s) represented by the equation (1) as the first probability. Various elements in the formula (1) will be described for each of the formulas (2) and (3) described later and the types of related web pages. The first probability is an example of the first index value.

Hereinafter, the details of the probability calculation process when the related web page is an encyclopedia web page will be described, and then the details of the probability calculation process when the related web page is a search result web page will be described.

<1-4. If the related web page is an encyclopedia web page>
The first probability calculation unit 113 calculates the probability that the text data indicating the entity associated with the target text data is included as the text (anchor text) indicating the link destination among the text data included in the encyclopedia web page. .. The related web page contains text data (illustrated text TX12) indicating the target text data (in this example, "3000 hits" (illustrated text TX11) and the associated entity (in this example, "baseball player A")). ~ 14) is included. Further, among the text TX12 ~ 14, the text TX12 is an anchor text. Further, the link of another web page given to the text TX12 is included in the target text data in the knowledge graph D1. The location information associated with the associated entity (in this example, "baseball player A").

Here, the first probability calculation unit 113 obtains the number of target texts in all the encyclopedia web pages including the target text data by the number of target text data which is the anchor text by the following equation (2). Calculate the probability of being generated (hereinafter referred to as the anchor text probability). As in equation (2) indicates the anchor text probability.

In addition, the first probability calculation unit 113 uses the following formula (3) to specify the web page of the entity whose link destination of the anchor text corresponds to the location information associated with the target text data in the knowledge graph D1 (hereinafter, specified. Web page) (hereinafter referred to as the first entity probability) is calculated. The e in the equation (3) indicates the first entity probability.

When the related web page collected by the related page collecting unit 112 is only the web page shown in FIG. 7, the first probability calculation unit 113 uses the target text data among the target text data included in all the related web pages. The probability that is an anchor text (that is, the anchor text probability) is calculated as "1/3". This is because, in the target text data shown in the figure, only one link is given (even if two or more links are given, it may be counted as one). Further, the first probability calculation unit 113 calculates the probability (first entity probability) that the link destination of the anchor text is a specific web page of the entity associated with the target text data as "1".

<1-5. If the related web page is a search result web page>
The first probability calculation unit 113 calculates the probability that the link destination including the entity associated with the target text data included in the search result web page as text is selected. The target text data is "3000 hits" (text TX21 in the figure), and the entity associated with the target text data is "baseball player A". In the following description, the text related to the entity associated with the target text data will be referred to as the target entity text data. In the example of FIG. 8, the related web page includes text data (texts TX22 to 24 shown) to which a link of the related web page of the target text data is added. Of these, the text TX22 contains the target entity text data (in this example, "baseball player A"). The link destination given to the text TX 22 is, for example, the location information associated with the entity (in this example, “baseball player A”) associated with the target text data in the knowledge graph D1. That is, the link of another web page of the text TX22 is a specific web page, for example, an encyclopedia web page. It should be noted that the search result web page may include text TX25, which is text without a link and indicates an entity.

The first probability calculation unit 113 is the number of times the link destination including the target entity text data is clicked with respect to the number of times each link destination included in all the search result web pages is clicked by the equation (2). Calculate the probability obtained as a ratio (hereinafter, entity text click probability). Note that "click" is an example of selection, and may be selected by touch, tap, or other modes. In the following description, it is collectively referred to as a click. As in the equation (2) indicates the entity text click probability.

Further, the first probability calculation unit 113 calculates the probability that the link destination including the target entity text data is a specific web page related to the entity (hereinafter, the second entity probability) by the formula (3). E in the equation (3) indicates the second entity probability.

<1-6. When the same entity pointing to different things is included>
Here, an entity may refer to different things even if they have the same wording. FIG. 9 is a diagram showing an example of linking to a web page of a different entity by the same wording. In the example shown in FIG. 9, the text TX22 indicates a link destination of a web page (for example, FIG. 3) of "baseball player A". On the other hand, the text TX26 indicates a link destination of a web page of a player who is a baseball player with the same surname and the same name as "baseball player A" and belongs to a baseball team C different from "baseball player A". The first probability calculation unit 113 also includes the click of the text TX26 in the entity click probability in the equation (2).

FIG. 10 is a diagram showing an example of a web page relating to another entity linked from the wording indicating the entity. The web page of "baseball player A" shown in FIG. 10 is a baseball player different from "baseball player A" ("baseball player A" shown in FIG. 3) associated with the text data "3000 hits". Related data related to a certain "baseball player A" is described.

When calculating the first entity probability, the first probability calculation unit 113 does not count the web page as the web page of the entity when the link destination is the web page shown in FIG. Further, even when the first probability calculation unit 113 calculates the second entity probability, if the link destination is the web page shown in FIG. 10, the web page is not counted as the web page of the entity.

<1-7. Weighting for each related web page>
Further, the first probability calculation unit 113 applies the following equation (4) to the value obtained by multiplying the anchor text probability calculated when the related web page is the encyclopedia web page by the first entity probability. You may further multiply. Further, the first probability calculation unit 113 has the following equation (4) for the value obtained by multiplying the entity text click probability calculated when the related web page is the search result web page by the second entity probability. May be further multiplied. Then, the first probability calculation unit 113 calculates the first probability (score) by, for example, adding both to obtain a logarithm.

Equation (4) is the number of samples for which equations (2) and (3) are calculated using the encyclopedia web page, and the number of samples for which equations (2) and (3) are calculated using the search result web page. The ratio with and is shown.

The first probability calculation unit 113 further multiplies the ratio of the number of samples for which equations (2) and (3) are calculated using the encyclopedia web page by the value obtained by multiplying the anchor text probability and the first entity probability. , Multiply the ratio of the number of samples for which equations (2) and (3) are calculated using the search result web page to the value obtained by multiplying the entity text click probability and the second entity probability, and add both. Then, the score is weighted.

The first probability calculation unit 113 stores the calculated score in association with the content information D2. FIG. 11 is a diagram showing an example of the content information D2 and the score according to the present embodiment.

Returning to FIG. 1, the context collecting unit 114 collects the context from the knowledge graph D1 stored in the storage unit 120, the related web page collected by the related page collecting unit 112, and the like. A context is a character string indicating arbitrary information that characterizes an entity such as an entity's attributes and types. The context collecting unit 114 stores the collected context as the context information D3 in the storage unit 120.

FIG. 12 is a diagram illustrating an example of a context collection process by the context collection unit 114 according to the present embodiment. The context collecting unit 114 collects, for example, a character string indicating a property in the knowledge graph, a name of an adjacent entity, and the like as a context. In the example shown in FIG. 12, the context collecting unit 114 collects "Germany", "175 cm" and the like as the context associated with the entity "Scholar A".

FIG. 13 is a diagram illustrating another example of the context collection process by the context collection unit 114 according to the present embodiment. The context collecting unit 114 collects, for example, a character string indicating the attribute of the object displayed after the character string of the object (entity) in the encyclopedia web page as a context. In the example shown in FIG. 13, the context collecting unit 114 collects "physics" described after the character string of the object "scholar A" as a context. FIG. 14 is a diagram showing an example of the context information D3 according to the present embodiment. In the example shown in FIG. 14, "physics", "Germany", "175 cm" and the like are associated with the entity "scholar A" as a context.

The second probability calculation unit 115 calculates the probability (P (context | entity)) (hereinafter, the second probability) indicating the degree of relevance to which each context is associated when a certain entity is given. The second probability calculation unit 115 generates a virtual document including frequency information of each context for each entity based on, for example, a click log of a related page, a search log of a search engine, and the like. Further, the second probability calculation unit 115 calculates the second probability by applying an algorithm (see Non-Patent Document 1) such as Latent Dirichlet Allocation (LDA) to the generated virtual document. That is, the second probability calculation unit 115 is based on the group information generated by the method of optimizing the feature amount when the entity and the word that can be the context are divided into virtual groups, and the given frequency information. Then, the second probability is calculated. The second probability is an example of the second index value.

FIG. 15 is a diagram illustrating an example of a context frequency information calculation process by the second probability calculation unit 115 according to the present embodiment. Here, the second probability calculation unit 115 is a query used by the user who clicks the related page (for example, the encyclopedia web page) of the entity "Scholar A" to display the search result page which is the transition source. Of these, the frequency information of each context is calculated for each entity based on the appearance frequency of words (context terms) other than "Scholar A". In the example shown in FIG. 15, for example, the frequency of "physics" having a high frequency of appearance is calculated as "120". On the other hand, the frequency of "height", which has a low appearance frequency, is calculated as "10". As shown in FIG. 14, the second probability calculation unit 115 stores the calculated frequency information in the storage unit 120 in association with the context information D3. Next, the second probability calculation unit 115 generates a virtual document by aggregating the contexts including these frequency information for each entity.

The second probability calculation unit 115 calculates the second probability by applying, for example, LDA to the generated virtual document. An entity corresponds to a virtual document, and the frequency information of the context for the entity corresponds to the phrase in the document.

Returning to FIG. 1, the location information acquisition unit 116 determines the set of the entity name, the entity, and the context corresponding to the query received from the web server 300. The location information acquisition unit 116 acquires the location information associated with the determined entity based on the knowledge graph D1. After that, the location information acquisition unit 116 transmits the acquired location information to the web server 300.

<1-8. Online operation of Knowledge Data Server 100>
FIG. 16 is a flow chart showing an example of online operation of the knowledge data server 100 of the present embodiment. First, the location information acquisition unit 116 receives a query from the web server 300 (S10). FIG. 17 is a diagram illustrating an online operation of the knowledge data server 100 of the present embodiment. The example shown in FIG. 17 shows an example of receiving a query containing two words, "Scholar A" and "Physics".

Next, the location information acquisition unit 116 reads out the first probability associated with the text data matching each of the words included in the received query in the content information D2 (S20). At this time, not only when the query and the text data are exactly matched, but also when there is a partial match or a difference to the extent that they are considered to be semantically identical, it may be regarded as the corresponding text data.

In the example shown in FIG. 17, as the entity named "Scholar A", there are a human entity EN1, a two-person entertainer entity EN2, and a racehorse entity EN3, and two words "scholar". Each of "A" and "Physics" shows an example associated with these three entities. In this case, the location information acquisition unit 116 has a probability that "scholar A" is associated with the entity EN1 that is a person, a probability that "scholar A" is associated with the entity EN2 that is a duo of entertainers, and "scholar A". The probability that "physics" is associated with the race horse entity EN3, the probability that "physics" is associated with the human entity EN1, the probability that "physics" is associated with the celebrity entity EN2, and the "physics". Reads a total of 6 probabilities with the probabilities that "" is associated with the race horse entity EN3.

Next, the location information acquisition unit 116 obtains a second probability from the context information D3 that each of the above three entities is associated with each of the two words "scholar A" and "physics" as contexts. Read (S30). In the example shown in FIG. 17, the location information acquisition unit 116 has a probability that the human entity EN1 is associated with the context "scholar A", a probability that the human entity EN1 is associated with the context "physics", and an entertainer. The probability that the two-person entity EN2 is associated with the context "Scholar A", the probability that the entertainer's two-person entity EN2 is associated with the context "physics", and the probability that the racehorse entity EN3 is associated with the context "Scholar A". , And the probability that the race horse entity EN3 is associated with the context "physics", for a total of six probabilities.

Next, the location information acquisition unit 116 multiplies the six first probabilities read in S20 and the six second probabilities read in S30 to each other (S40). In the example shown in FIG. 17, the location information acquisition unit 116 calculates a total of 36 product values.

Next, the location information acquisition unit 116 determines the set of the entity name, the entity, and the context having the largest value among the calculated multiplication values as an appropriate set as the search intention of the query input by the user (S50). ). That is, the location information acquisition unit 116 determines that the larger the multiplication value, the higher the degree of relevance of the word to the entity. For example, in the example shown in FIG. 17, the location information acquisition unit 116 uses a set whose entity name is "scholar A", whose entity is "person", and whose context is "physics" as the search intent of the query. Determine as an appropriate pair. The location information acquisition unit 116 acquires the location information of the entity included in the determined set from the knowledge graph D1 and transmits it to the web server 300. This completes the processing of this flowchart.

<1-9. First offline operation of Knowledge Data Server 100>
Hereinafter, the first offline operation of the knowledge data server 100 will be described. In the first offline operation, the first probability is calculated. FIG. 18 is a flow chart showing an example of an operation of calculating the first probability of the knowledge data server 100 of the present embodiment. The selection unit 111 selects the target text data to be calculated from the content information D2 (S110). Next, the related page collecting unit 112 collects related web pages (encyclopedia web page and search result web page) based on the target text data (S120). Next, the first probability calculation unit 113 calculates the anchor text probability and the entity text click probability (S130).

Next, the first probability calculation unit 113 calculates the first entity probability and the second entity probability (S140). Then, the first probability calculation unit 113 multiplies the anchor text probability based on the encyclopedia web page, the first entity probability, and the first content ratio, the entity text click probability based on the search result web page, and the second entity. A value multiplied by the probability and the value obtained by multiplying the second content ratio is calculated as the first probability (S150), and the calculated first probability is associated with the content information D2 and stored in the storage unit 120 (S160).

<1-10. Second offline operation of Knowledge Data Server 100>
Hereinafter, the second offline operation of the knowledge data server 100 will be described. In the second offline operation, the second probability is calculated. FIG. 19 is a flow chart showing an example of an operation of calculating the second probability of the knowledge data server 100 of the present embodiment. The selection unit 111 selects the target text data to be calculated from the content information D2 (S210). The related page collection unit 112 collects related web pages based on the target text data. Next, the context collecting unit 114 collects the context based on the related web page collected by the related page collecting unit 112 (S220).

Next, the second probability calculation unit 115 calculates the frequency information of each context for each entity (S230), and generates a virtual document including the calculated frequency information. Next, the second probability calculation unit 115 calculates the second probability by applying an algorithm such as LDA to the generated virtual document (S240).

FIG. 20 is a flow chart showing an example of a detailed operation for calculating the second probability of the knowledge data server 100 of the present embodiment. In LDA, the characteristics of a sentence are analyzed by assuming that the document is composed of a plurality of topics (for example, 1000 topics) (groups). A large number of patterns of multiple topics (hereinafter referred to as topic patterns) are defined. That is, one topic pattern includes a plurality of topics, and many topic patterns having different ways of dividing the plurality of topics are defined. In this embodiment, the entity corresponds to the document, and the frequency of the context corresponding to the entity corresponds to the frequency of the phrase in the document. First, the second probability calculation unit 115 selects one pattern from a large number of topic patterns (S241).

Next, the second probability calculation unit 115 selects a candidate for a set of the entity and the context to be calculated (S242). Next, the second probability calculation unit 115 calculates the second probability (P (context | entity)) (S243). For example, the second probability calculation unit 115 obtains the product of the probability that the context belongs to the topic and the probability that the topic belongs to the entity (by content) based on the following equation (5). The second probability calculation unit 115 learns the distribution so that the value of the product of these probabilities is maximized.

Next, the second probability calculation unit 115 determines whether or not the calculation of the second probability is completed for the candidates of the set of all the entities and the contexts (S244). When the second probability calculation unit 115 determines that the calculation of the second probability has not been completed for the candidates of the set of all the entities and the context, the set of the entity and the context that has not performed the calculation process of the second probability. Is selected, and the processing after S242 is repeated. On the other hand, when the second probability calculation unit 115 determines that the calculation of the second probability is completed for the candidates of the set of all the entities and the contexts, the second probability calculation unit 115 performs the processing of adding the calculated second probabilities to the processing target. Calculate the score for the topic pattern.

Next, the second probability calculation unit 115 determines whether or not the processing for all the topic patterns is completed (S245). When the second probability calculation unit 115 determines that the processing for all the topic patterns is not completed, the second probability calculation unit 115 selects one topic pattern that has not been processed, and repeats the processing after S242.

On the other hand, when it is determined that the processing for all the topic patterns is completed, the second probability calculation unit 115 determines the topic pattern having the highest calculated score as the topic pattern to be adopted (S246).

Next, the second probability calculation unit 115 determines the second probability calculated for the topic pattern determined as described above as the second probability (S247). The second probability calculation unit 115 does not have to calculate the second probability for all the candidates of the set of the entity and the context. For example, if the candidate pair of entity and context contains an inappropriate expression, it may be excluded. In addition, the second probability calculation unit 115 may perform exclusion processing according to additional conditions. For example, the second probability calculation unit 115 may exclude a context that is not included in the summary text of the encyclopedia web page corresponding to the entity. The data may be excluded after the probability calculation is completed, instead of being excluded before the calculation of the second probability. This is because even an inappropriate expression is useful for calculating the probability in LDA. This completes the processing of this flowchart.

Returning to FIG. 19, the second probability calculation unit 115 stores the determined second probability in the storage unit 120 in association with the context information D3 (S250). The second probability calculation unit 115 may store the determined second probability in an inverted index such as Solr. By throwing an OR search query to this Solr, it is possible to acquire all the candidates with one inquiry without making inquiries many times. This completes the processing of this flowchart.

As described above, the knowledge data server 100 of the present embodiment calculates the first probability of calculating the first index value indicating the degree of relevance between each of the two or more words input as the search query and the entity. The second probability calculation unit 115 (second index value) for calculating the second index value indicating the degree of association between the unit 113 (first index value calculation unit), the entity including the word as a context, and each of the words. Based on the calculation unit), the first index value calculated by the first probability calculation unit 113, and the second index value calculated by the second probability calculation unit 115, the word is highly related. It is provided with a location information acquisition unit 116 (determination unit) for determining an entity. As a result, the knowledge data server 100 of the present embodiment can suppress the output of inappropriate information together with the search result.

Although the embodiments for carrying out the present invention have been described above using the embodiments, the present invention is not limited to these embodiments, and various modifications and substitutions are made without departing from the gist of the present invention. Can be added.

100 ... Knowledge data server 110 ... Control unit 111 ... Selection unit 112 ... Related page collection unit 113 ... First probability calculation unit 114 ... Context collection unit 115 ... Second probability calculation unit 116 ... Location information acquisition unit 117 ... Communication I / F
120 ... Storage unit 200 ... Terminal device 210 ... Display unit 211 ... Query input area 220 ... Knowledge panel 221 ... Entity 222 ... Image 223 ... Related information 230 ... Search result 300 ... Web server D1 ... Knowledge graph D2 ... Content information D3 ... Context information

Claims (10)

Each of the two or more words entered as a search query, the first index value calculation unit that calculates the first index value indicating the degree of relevance to the entity, and
An entity that includes the word as a context, a second index value calculation unit that calculates a second index value indicating the degree of association with each of the words, and a second index value calculation unit.
Judgment to determine an entity highly related to the word based on the first index value calculated by the first index value calculation unit and the second index value calculated by the second index value calculation unit. Equipped with a department,
The determination unit has a word corresponding to the name of the entity having a high relationship with the word among the two or more words, a word corresponding to the context among the two or more words, and a word having a high relationship with the word. Determine the pair with the entity,
Information processing equipment. The first index value is the probability that each of the words assumed as the name of the entity is associated with each of the plurality of the entities.
The second index value is the probability that each of the entities including the plurality of said words as a context is associated with each of the said words assumed as the context.
The information processing apparatus according to claim 1. The determination unit determines that the larger the product of the first index value and the second index value, the higher the relationship with the word of the entity .
The information processing apparatus according to claim 1 or 2. The second index value calculation unit is based on the group information generated by the method of optimizing the feature amount when the entity and the word that can be the context are divided into virtual groups, and the given frequency information. , Calculate the second index value,
The information processing apparatus according to any one of claims 1 to 3. Further provided with a frequency information calculation unit that calculates the frequency information based on the click log of the page associated with the entity that includes the word as a context .
The information processing apparatus according to claim 4. Two entered as a search query based on the group information generated by the method of optimizing the features when the words that can be entities and contexts are divided into virtual groups, and the given frequency information. It is equipped with a calculation unit that calculates an index value indicating the degree of relevance between the entity including the above words as a context and each of the words.
The feature amount is a product of the probability that the context belongs to the group and the probability that the group belongs to an entity that includes the word as a context .
Based on the index value calculated by the calculation unit, the word corresponding to the name of the entity highly related to the word among the two or more words, and the context among the two or more words. Further provided with a determination unit for determining a pair of a word and an entity closely related to the word.
Information processing equipment. The computer
Calculate the first index value indicating the degree of relevance between each of the two or more words entered as a search query and the entity.
A second index value indicating the degree of association between the entity including the word as a context and each of the words is calculated.
It is an information processing method for determining an entity highly related to the word based on the first index value and the second index value.
Determining an entity that is highly related to the word is a word that corresponds to the name of the entity that is highly related to the word among the two or more words, and a word that corresponds to the context among the two or more words. And to determine the pair of the entity that is closely related to the word,
Information processing method. The computer
Two entered as a search query based on the group information generated by the method of optimizing the features when the words that can be entities and contexts are divided into virtual groups, and the given frequency information. An index value indicating the degree of relevance between the entity containing the above words as a context and each of the words is calculated.
The feature amount is an information processing method that is a product of the probability that the context belongs to the group and the probability that the group belongs to an entity that includes the word as a context .
Further, based on the calculated index value, the word corresponding to the name of the entity highly related to the word among the two or more words, and the word corresponding to the context among the two or more words. , Determine the pair of the entity that is highly related to the word,
Information processing method. On the computer
Have each of the two or more words entered as a search query calculate the first index value that indicates the degree of relevance to the entity.
A second index value indicating the degree of relevance between the entity including the word as a context and each of the words is calculated.
A program for determining an entity highly related to the word based on the first index value and the second index value.
To determine an entity that is highly related to the word is a word that corresponds to the name of the entity that is highly related to the word among the two or more words, and a word that corresponds to the context among the two or more words. And to determine the pair of the entity closely related to the word,
program. On the computer
Two entered as a search query based on the group information generated by the method of optimizing the features when the words that can be entities and contexts are divided into virtual groups, and the given frequency information. An index value indicating the degree of relevance between the entity including the above words as a context and each of the words is calculated.
The feature amount is a program that is a product of the probability that the context belongs to the group and the probability that the group belongs to an entity that includes the word as a context .
Further, based on the calculated index value, the word corresponding to the name of the entity highly related to the word among the two or more words, and the word corresponding to the context among the two or more words. , To determine the pair of the entity that is highly related to the word,
program.

Images