JP6739361B2 - Information providing device, information providing method, and program - Google Patents

Description

The present invention relates to an information providing device, an information providing method, and a program.

2. Description of the Related Art Conventionally, in web search, research has been conducted on a technique for analyzing the content of a question sentence input by a user and obtaining a search result including an answer requested by the user. For example, by analyzing the question text entered by the user, whether the question text asks for "personal name", "place name", or "date" as the answer. Research on techniques for determining the label, which is the question type, and improving the accuracy of search results is underway (see, for example, Patent Documents 1 and 2).

JP 2012-68687 A JP, 2010-198189, A

In the case of determining the question content label in the conventional technique, it is necessary to rule the relationship between the reference question content pattern and the label in advance. For example, a label "Person (person name)" is given to the question content pattern "Who is ..." and a "Date (date )” is required to be defined in advance.

In many cases, the rule formation as described above is performed manually, and the workload required for the rule formation may increase. In addition, in order to change the label such as increasing the label definition, it was necessary to reconstruct the rule. Furthermore, since the accuracy of such rules depends on the skill of the worker who makes the rules, the accuracy of the search results can be improved if the question sentence pattern or label settings are not appropriate. There was a problem that I could not.

The present invention has been made in view of such circumstances, and an information providing device, an information providing method, which can improve the accuracy of search results by automating and optimizing the labeling of question contents. And one of the purposes is to provide a program.

One aspect of the present invention uses a first search query that is a set of a first character string that is a keyword extracted from the first question content and a second character string that is an answer content to the first question content. A first search unit that performs a network search to obtain a first document as a search result, and a third morpheme that is before and after the second character string included in the first document obtained by the first search unit. An information providing apparatus, comprising: an assigning unit that assigns a first label indicating a question type to the first question content based on a character string.

According to one aspect of the present invention, the accuracy of search results can be improved by automating and optimizing the labeling of question contents.

It is a block diagram of the information providing system 1. It is a figure which shows the function structure of the information provision apparatus 5. It is a figure which shows the function structure of the determination part 14. FIG. 6 is a diagram showing an example of a record in which an answer content AL is associated with a question content QL stored in a storage unit 20. FIG. 9 is a diagram showing an example of a record after a label L is added to the question content QL stored in the storage unit 20. It is a flowchart which shows the process of the information provision apparatus 5 in a learning stage. It is a flowchart which shows the process of the information provision apparatus 5 in an operation stage.

Hereinafter, embodiments of an information providing apparatus, an information providing method, and a program of the present invention will be described with reference to the drawings. The information providing device performs a network search using a search query that is a combination of a keyword extracted from the question content and the answer content to obtain a document that is a search result, and a character string of the answer content included in this document. It is a device that adds a label indicating a question type to the question content based on the morphemes before and after. Further, the information providing apparatus learns the relationship between the question content and the label in advance and determines the label associated with the question content requesting the information provision.

FIG. 1 is a configuration diagram of the information providing system 1. The information providing system 1 realizes an interactive information providing service. For example, the information providing system 1 realizes a question-and-answer type information providing service that provides the answer content "Tokyo" to the question content "where is the capital of Japan?" from the user.

The information providing system 1 includes, for example, one or more terminal devices 3, one or more information providing devices 5, and one or more search servers 7. The terminal device 3, the information providing device 5, and the search server 7 are connected to each other via a network NW, and communicate with each other via this network NW. The network NW includes, for example, a WAN (Wide Area Network), a LAN (Local Area Network), the Internet, a dedicated line, a wireless base station, a provider, and the like.

[Terminal device]
The terminal device 3 is operated by a user who uses an interactive information providing service. The terminal device 3 is, for example, a personal computer, a mobile phone such as a smartphone, a tablet terminal, or a computer device such as a PDA (Personal Digital Assistant). The terminal device 3 transmits the question content Q (second question content) requesting the information provision to the information providing apparatus 5 and receives the answer content A to the question content Q from the information providing apparatus 5 based on the operation of the user. ..

[Information providing device]
The information providing device 5 is a computer device that transmits the answer content A to the question content Q received from the terminal device 3 to the terminal device 3. FIG. 2 is a diagram showing a functional configuration of the information providing device 5. The information providing device 5 includes, for example, a communication unit 10 (reception unit), a question content search unit 12 (second search unit), a determination unit 14, an answer determination unit 16 (determination unit), and a storage unit 18. Prepare Each functional unit included in the information providing device 5 may be distributed to a plurality of devices. For example, the determination unit 14 and the other functional units may be realized by a device separate from each other. The storage unit 18 may be a storage device such as NAS (Network Attached Storage).

The question content search unit 12, the determination unit 14, and the answer determination unit 16 are realized by a processor such as a CPU (Central Processing Unit) executing a program (software) stored in the storage unit 18, for example. The program may be downloaded from an application server via the network NW, or may be preinstalled in the information providing device 5, for example. Further, these functional units may be realized by hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or the cooperation of software and hardware. May be realized by. The storage unit 18 is realized by, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a HDD (Hard Disk Drive), a flash memory, or a hybrid storage device in which a plurality of these are combined.

The communication unit 10 receives the question content Q from the terminal device 3, and outputs the received question content Q to the question content search unit 12 and the determination unit 14. The communication unit 10 also transmits the answer content A determined by the answer determination unit 16 to the terminal device 3.

The question content search unit 12 performs a network search using the question content Q input from the communication unit 10 as a search query (second search query). For example, the question content search unit 12 transmits a search query, which is the question content Q input from the communication unit 10, to the search server 7, and the search server 7 returns at least 1 page of the URL of the content associated with the search query. One is acquired, and the acquired page is output to the answer determination unit 16 as the search result D. The acquired page is a document (second document) that is text data described in a markup language such as HTML (Hyper Text Markup Language). In addition to the question content Q, the question content search unit 12 may perform a query expansion search using a term related to the question content Q. Further, the question content search unit 12 may acquire a part of the page of the URL of the content associated with the search query from the search server 7.

The determination unit 14 determines the label L, which is the question type of the question content Q input from the communication unit 10. As the label L, for example, an organization name (Organization), a person name (Person), a place name (Location), a date expression (Date), a time expression (Time), an amount expression (Money), a percentage expression (Percent), and a unique item Name (Artifact) etc. are included. The definition of the label is arbitrarily set by the administrator of the information providing device 5 or the like. The determination unit 14 outputs the determined label L to the answer determination unit 16.

In the determination processing of the label L in the determination unit 14, a recurrent neural network (RNN), a multilayered neural network (Deep Neural Network: DNN), a convolutional neural network (CNN), or the like is used. Adopt deep learning technology.

FIG. 3 is a diagram showing a functional configuration of the determination unit 14. The determination unit 14 includes, for example, a storage unit 20, a reading unit 22, a keyword search unit 24 (first search unit), a label assignment unit 26 (assignment unit), and a label determination unit 28 (determination unit). .. The storage unit 20, the reading unit 22, the keyword searching unit 24, and the label assigning unit 26 are functional units that operate at the learning stage of the label determining unit 28.

The storage unit 20 stores in advance a plurality of records in which the answer content AL is associated with the question content QL (first question content). These plurality of records are records that are the sources of the learning data that the label determination unit 28 learns. These plurality of records may be provided from an external database that provides learning data, or may be automatically generated from content on the Internet. Further, these plural records may be manually registered by the administrator of the information providing device 5, or plural question contents may be mechanically generated from the answer contents. A label L is attached to the question content QL in each of the plurality of records by the label attaching unit 26. The record to which the label L has been attached by the label attaching unit 26 becomes the learning data to be learned by the label determining unit 28.

FIG. 4 is a diagram showing an example of a record in which the answer content AL is associated with the question content QL stored in advance in the storage unit 20. FIG. 5 is a diagram showing an example of a record after the label L is attached to the question content QL stored in the storage unit 20.

In the example shown in FIG. 4, the storage unit 20 stores a record in which the answer content AL of “Tokyo” is associated with the question content QL of “where is the capital of Japan?”. Here, for example, the label L is not given to these records that are provided from an external database and stored in the storage unit 20 in advance. On the other hand, in the example shown in FIG. 5, the label assigning unit 26 assigns the label L of “place name” to the question content QL “where is the capital of Japan?” by the label assigning unit 26. Recorded records are stored.

The storage unit 20 is realized by, for example, a RAM, a ROM, a HDD, a flash memory, or a hybrid storage device in which a plurality of these are combined. The storage unit 20 may be the same device as the storage unit 18.

The reading unit 22 reads a set of question contents QL and answer contents AL as shown in FIG. 4 from the storage unit 20. The reading unit 22 performs morphological analysis of the read question content QL to extract at least one keyword K (first character string), and a combination of the extracted keyword K and the read answer content AL (second character string). Is output to the keyword search unit 24. Further, the reading unit 22 outputs the read response content AL to the label assigning unit 26.

The keyword search unit 24 performs a network search by using the combination of the keyword K and the answer content AL input from the reading unit 22 as a search query (first search query). For example, the keyword search unit 24 transmits a search query, which is a set of the keyword K and the answer content AL input from the reading unit 22, to the search server 7, and the search server 7 sends the URL of the content associated with the search query. At least one page is acquired, and the acquired page is output to the labeling unit 26 as the search result DL. The acquired page is, for example, a document (first document) which is text data described in a markup language such as HTML. The keyword search unit 24 may acquire a part of the URL page of the content associated with the search query from the search server 7.

The label assigning unit 26 assigns a label L (first label) to the question content QL by applying a unique expression extraction technique by morphological analysis. Here, the proper expression extraction technique is a technique that determines the type of a term according to the co-occurrence of a given term by morphological analysis, whether the given term is a suffix or a prefix. .. For example, the labeling unit 26 morphologically analyzes the text of the search result D, and associates the answer content AL with the character string (third character string) that is a morpheme before and after the character string of the answer content AL. A label L is given to the question content QL.

For example, when the morpheme before the character string of the response content AL is “corporation...”, “corporation corporation...”, etc., the label assigning unit 26 assigns “organization name” as the label L, and ..", "prefectural office location...", etc., "place name" is given as the label L. When the morpheme after the character string of the answer content AL is "... prefecture", "... departure", etc., the label assigning unit 26 assigns "place name" as the label L and "... house". , "...", "person name" is given as the label L.

In the learning stage, the label determination unit 28 uses the record after the label L is attached to the question content QL as shown in FIG. 5 as learning data, and determines the relationship between the question content QL and the label L. learn. Further, the label determination unit 28 determines the label L (second label) of the question content Q (question content Q from the user) input from the communication unit 10 in the operation stage, and the determined label L is the answer determination unit. Output to 16.

The answer determination unit 16 determines the answer content A for the question content Q based on the label L input from the determination unit 14 and the search result D input from the question content search unit 12. For example, the answer determination unit 16 performs a morphological analysis of the document of the search result D, extracts a term having high relevance to the label L, and determines the extracted term as the answer content A. In addition, the answer determination unit 16 may perform morphological analysis of the search result D to filter terms that are less relevant to the label L. In addition, the answer determination unit 16 performs scoring based on the appearance frequency of the extracted term, the search order of the search result D that is the extraction source, the positional relationship of the extracted term in the document of the search result D, and the like. A term that is judged to be highly valid as the answer content A for the question content Q may be determined as the answer content A. The answer determination unit 16 transmits the determined answer content A to the terminal device 3 via the communication unit 10.

The storage unit 18 stores, for example, the question content Q received by the communication unit 10 from the terminal device 3, the URL of the content and the page received by the question content search unit 12 from the search server 7, and the like.

[Search Server]
The search server 7 is a computer device that receives a search query from the information providing device 5 and provides a search result based on the received search query. The search server 7 includes a search database (not shown) that associates the search query with the URL of the content and the page of the URL or a part thereof. When receiving the search query from the information providing device 5, the search server 7 extracts the URL of the content associated with the search query and the page of the URL or a part thereof from the search database. The search server 7 transmits the URL of the content extracted from the search database and the page of the URL or a part thereof to the information providing device 5.

[Processing of information providing device (learning stage)]
Next, the operation of the information providing device 5 in the learning stage will be described with reference to FIG. FIG. 6 is a flowchart showing the processing of the information providing device 5 in the learning stage.

First, the reading unit 22 included in the determination unit 14 reads one record of the set of question content QL and answer content AL as shown in FIG. 4 from the storage unit 20 (S101). The reading unit 22 extracts at least one keyword K from the read question content QL, and outputs a set of the extracted keyword K and the read answer content AL to the keyword searching unit 24. Further, the reading unit 22 outputs the read response content AL to the label assigning unit 26.

Next, the keyword search unit 24 performs a search using the keyword K and the answer content AL input from the reading unit 22 (S103). For example, the keyword search unit 24 transmits a search query, which is a set of the keyword K and the answer content AL input from the reading unit 22, to the search server 7, and the search server 7 sends the URL of the content associated with the search query. At least one document that is a page is acquired, and the acquired document is output to the label assigning unit 26 as a search result DL.

Next, the label assigning unit 26 morphologically analyzes the text of the search result D, and associates the answer content AL with the character string (third character string) that is a morpheme before and after the character string of the answer content AL. A label is attached to the question content QL that has been asked (S105). For example, the label assigning unit 26 assigns a label to the question content QL stored in the storage unit 20 by associating the label L with the question content QL and storing the label L in the storage unit 20.

Next, the reading unit 22 determines whether reading of all the records stored in the storage unit 20 has been completed (S107). When the reading unit 22 determines that reading of all the records has not been completed, the reading unit 22 reads one record of the unread question content QL and answer content AL from the storage unit 20, and reads the above label. L is given.

On the other hand, when the reading unit 22 determines that all the records have been read, the label determining unit 28 performs a learning process for learning the relationship between the question content QL and the label L (S109). The label determination unit 28 learns the relationship between the question content QL and the label L, with the record after the label L is added to the question content QL as shown in FIG.

For example, the label determination unit 28 performs a morpheme analysis on the question content QL "where is the capital of Japan?" and performs the morpheme "Japan/no/capital/wa/where/is/ka/?/". , Each morpheme is converted into a fixed-length vector, this vector is inserted into each layer of the recurrent neural network in order, weighting is performed using a softmax function, etc., association with the label is performed, and the question content QL And the label L are learned. By performing such learning, the label determination unit 28 can determine the label L to be associated with the question content Q when the question content Q requesting the information provision is received. With the above, the information providing apparatus 5 ends the processing of this flowchart.

[Processing of information providing device (operation stage)]
Next, the operation of the information providing device 5 in the operation stage will be described with reference to FIG. FIG. 7 is a flowchart showing the processing of the information providing device 5 in the operation stage.

First, the communication unit 10 acquires the question content Q received from the terminal device 3 according to the user's operation (S201). The communication unit 10 outputs the acquired question content Q to the question content search unit 12 and the determination unit 14.

Next, the question content search unit 12 performs a search process using the question content Q input from the communication unit 10 as a search query (S203). For example, the question content search unit 12 transmits the search query, which is the question content Q input from the communication unit 10, to the search server 7, and the search server 7 displays the document that is the page of the URL of the content associated with the search query. At least one document is acquired, and the acquired document is output to the answer determination unit 16 as the search result D.

Next, in parallel with the search process by the question content search unit 12, or before or after the search process, the label determination unit 28 included in the determination unit 14 causes the question content Q input from the communication unit 10 to be input. A label L to be associated with is determined (S205). For example, the label determination unit 28 performs a morphological analysis of the question content Q and determines a label L to be associated with the question content Q using a recurrent neural network (RNN). The label determination unit 28 outputs the determined label L to the answer determination unit 16.

The answer determination unit 16 determines the answer content A for the question content Q based on the label L input from the label determination unit 28 and the search result D input from the question content search unit 12 (S207). For example, the answer determination unit 16 performs a morphological analysis of the document of the search result D, extracts a term having high relevance to the label L, and determines the extracted term as the answer content A. The answer determination unit 16 transmits the determined answer content A to the terminal device 3 via the communication unit 10. With the above, the information providing apparatus 5 ends the processing of this flowchart.

According to the embodiment described above, the search query (which is a set of the keyword K (first character string) extracted from the question content QL (first question content) stored in the storage unit 20 and the answer content QL ( It is included in the keyword search unit 24 (first search unit) that performs the network search using the first search query) to obtain the search result DL (first document), and the search result DL obtained by the keyword search unit 24. A label assigning unit 26 (assigning unit) that assigns a label L (first label) indicating a question type to the question content QL based on morphemes (third character string) before and after the answer content QL. This makes it possible to automate and optimize the labeling of the question content and improve the accuracy of search results.

According to the embodiment described above, since the labeling of question contents is automated, it is possible to reduce the work load associated with labeling and to optimize label setting. Further, even if it is desired to change the label such as increasing the definition of the label, the learning process can be performed again, so that it is not necessary to reconstruct the rule as in the conventional case.

In the above embodiment, an example is described in which the information providing device 5 receives the question content Q from the terminal device 3 connected via the network NW and transmits the answer content A to the question content Q to the terminal device 3. did. However, the function of the information providing device 5 may be implemented in the terminal device 3.

As described above, the embodiments for carrying out the present invention have been described using the embodiments, but the present invention is not limited to such embodiments, and various modifications and substitutions are made within the scope not departing from the gist of the present invention. Can be added.

1... Information providing system, 3... Terminal device, 5... Information providing device, 7... Search server, 10... Communication unit, 12... Question content search unit, 14... Judgment unit, 16... Answer determination unit, 18... Storage unit, 20... Storage unit, 22... Reading unit, 24... Keyword search unit, 26... Label assigning unit, 28... Label determination unit, NW... Network

Claims (7)

A network search is performed using a first search query that is a set of a first character string that is a keyword extracted from the first question content and a second character string that is an answer content to the first question content, and a search result A first search unit for obtaining the first document of
A first type indicating a question type for the first question content based on a third character string that is a morpheme before and after the second character string included in the first document acquired by the first search unit. An assigning part that assigns a label,
An information providing device including. A determination unit that further learns the relationship between the first question content and the first label assigned by the assignment unit in advance and determines the second label associated with the second question content that requests information provision is further provided. Prepare,
The information providing device according to claim 1. A receiving unit that receives the second question content from the terminal device;
A second search unit that performs a network search using the second search query, which is the content of the second question, and obtains a second document as a search result;
Further comprising a determination unit that determines the content of the answer to the second question content based on the second label determined by the determination unit and the second document acquired by the second search unit,
The information providing device according to claim 2. The determination unit performs a morphological analysis of the second document, extracts a term associated with the second label determined by the determination unit, and determines the extracted term as an answer content to the second question content. To do
The information providing device according to claim 3. The information providing device receives the second question content from a terminal device connected via a network and transmits an answer content to the second question content to the terminal device.
The information providing device according to claim 2. Computer
A network search is performed using a first search query that is a set of a first character string that is a keyword extracted from the first question content and a second character string that is an answer content to the first question content, and a search result Get the first document of
Assigning a first label indicating a question type to the first question content based on a third character string that is a morpheme before and after the second character string included in the acquired first document,
Information provision method. On the computer,
A network search is performed using a first search query that is a set of a first character string that is a keyword extracted from the first question content and a second character string that is an answer content to the first question content, and a search result To get the first document of
A first label indicating a question type is given to the first question content based on a third character string that is a morpheme before and after the second character string included in the acquired first document,
program.

Images