KR20190074828A - Automatic Question-Answering system having mutiple Question-Answering modules - Google Patents

Abstract

Provided is an automatic question answering system capable of quickly and accurately answering a question. According to one aspect of the inventive concept of the present invention, the automatic question answering system comprises: a user interface receiving a question of a user in a form of a natural language from the user through a network, and transmitting an answer; at least one module server having a plurality of question answering (QA) modules capable of answering questions; a question analysis unit performing the analysis on the natural language of the question; a module selection execution unit selecting some of the plurality of QA modules based on an analysis result of the natural language of the question; and a question answering result integrating unit providing an answer for the question to the user through the user interface. The module selection execution unit comprises: a question answering action predicting unit receiving the analysis result of the natural language of the question and predicting, through learning, question action conformity for each of the plurality of QA modules with respect to the question; a question answering module selection unit selecting some of the plurality of QA modules with reference to a question answering coverage and the question action conformity set for each of the plurality of QA modules; and a question answering module execution unit providing the analysis result of the natural language of the question to some of the plurality of QA modules, which are selected at the question answering module selection unit, so that question answering is performed.

Description

[0001] The present invention relates to an automatic question-answering system having a plurality of query response modules,

The present invention relates to an automatic query response system, and more particularly to an automatic query response system having a plurality of QA (Question-Answering) modules.

The present invention is derived from research conducted and conducted by Saltlux Co., Ltd. as part of SW Technology Computing Industry Source Technology Development (SW) of Ministry of Science, Technology and Information. [Research period: 2017.03.01 ~ 2017.12.31] Research institute: Information and Communication Technology Promotion Center, Research title: WiseKB: Big data understanding based self-learning knowledge base and inference technology development, task number: 2013-0 -00109]

The automatic question-and-answer system is a system for responding to a user's query, and can analyze the user's query and provide a response to the user. On the other hand, in order to respond to various queries of the user, there is a demand for an automatic question and answer system having a plurality of QA modules capable of performing a query response independently or connected to a plurality of QA modules.

However, in the automatic question and answer system having a plurality of QA modules or connected to a plurality of QA modules, since each of the plurality of QA modules independently performs a query response, the speed of obtaining a response from all of the plurality of QA modules increases, There is a difficulty ranking in response.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an automatic question and answer system capable of quickly and accurately responding to queries.

According to an aspect of the present invention, there is provided an automatic query response system including: a user interface for receiving a user query in a natural language form through a network and transmitting a response; At least one module server having a plurality of QA modules capable of performing a query response; A query analyzer for performing a natural language analysis on the query; A module selection execution unit for selecting some of the plurality of QA modules based on a natural language analysis result of the query; And a query response result integration unit for providing the response to the query to the user through the user interface, wherein the module selection execution unit receives the natural language analysis result of the query, A query response operation predicting unit for predicting a query operation fitness of each QA module through learning; A query response module selection unit for selecting some of the plurality of QA modules with reference to the query response coverage set for each of the plurality of QA modules and the query operation suitability; And a query response module execution unit for providing a natural language analysis result of the query to some of the plurality of QA modules selected by the query response module selection unit to perform a query response.

The automatic query response system according to the present invention selects only a part of QA modules suitable for a query among a plurality of QA modules and performs a query response, so that the load is minimized and the time is shortened in both the query response execution and the response generation process, An accurate response can be provided to the user.

1 is a block diagram illustrating an automatic question-and-answer system in accordance with an exemplary embodiment of the present invention.
2 is a block diagram illustrating an operation of a query analyzing unit of an automatic question-answering system according to an exemplary embodiment of the present invention.
3 is a block diagram illustrating the operation of the module selection execution unit of the automatic question answering system according to an exemplary embodiment of the present invention.
4 is a block diagram for explaining the operation of the query response operation predicting unit and the query response module selecting unit of the automatic question and answer system according to the exemplary embodiment of the present invention.
5 is a block diagram illustrating an automated query response system in accordance with an exemplary embodiment of the present invention.
6 is a detailed block diagram illustrating a domain module server and a domain DB unit of the automatic question and answer system according to an exemplary embodiment of the present invention.
7 is a block diagram illustrating the plug-in module server of the automatic question-answering system according to an exemplary embodiment of the present invention in detail.
8 is a block diagram illustrating in detail an external service module sub and an external service / DB coupled thereto according to an exemplary embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Embodiments of the present invention are provided to more fully describe the present invention to those skilled in the art. The present invention is capable of various modifications and various forms, and specific embodiments are illustrated and described in detail in the drawings. It should be understood, however, that the invention is not intended to be limited to the particular forms disclosed, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for similar elements in describing each drawing. In the accompanying drawings, the dimensions of the structures are enlarged or reduced from the actual dimensions for the sake of clarity of the present invention.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprises", "having", and the like are used to specify that a feature, a number, a step, an operation, an element, a part or a combination thereof is described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be construed to have meanings consistent with the contextual meanings of the related art and are not to be construed as ideal or overly formal meanings as are expressly defined in the present application .

In the drawings and the description below, the components indicated or described as one block may be a hardware block or a software block. For example, each of the components may be an independent hardware block that sends and receives signals to each other, or may be a software block that executes on one processor.

In order to fully understand the structure and effects of the present invention, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a block diagram illustrating an automatic question-and-answer system in accordance with an exemplary embodiment of the present invention.

Referring to Figure 1, an automatic question and answer system 1 includes a user interface (UI) 100 that receives a query from a user 10 via a network 20 and sends a response. The network 20 may be a wired Internet service, a local area network (LAN), a wide area network (WAN), an intranet, a wireless Internet service, a mobile computing service, a wireless data communication service, a wireless Internet access service, And can transmit / receive data through wire / wireless. When the network 20 is connected to a smartphone or a tablet, the network 20 may be a wireless data communication service such as 3G, 4G, or 5G, a wireless LAN such as a Wi-Fi, Bluetooth, or the like.

The user interface 100 may provide an interface for accessing the automatic question and answer system 1 through a terminal or the like used by the user 10. The user 10 can send a query to the automatic question and answer system 1 via the user interface 100 and receive a response to the query provided by the automatic question and answer system 1 via the user interface 100 .

The automatic question and answer system 1 is connected to the user interface 100 and receives a query from the user 10 and determines a response and transmits a response to the user 10 through the user interface 100. [ 200). Also, it can control the components included in the automatic question and answer system 1 by the query response management unit 200.

The automatic question-answering system 1 includes a natural language understanding unit 300 for allowing a query response manager 200 to analyze a query of a natural language type, a query response manager 300 for referring to the natural language understanding unit 300, (500, 600, 700) having a plurality of modules (510, 520, 530, 540, 610, 620, 630, 710, 720, 730) ).

For example, the module sub 500, 600, 700 may include at least one of a domain module server 500, a plugin module server 600, and an external service module server 700. For example, the plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, 730 may include a plurality of domain modules 510, 520, 530, 540, a plurality of plug- , 630, and a plurality of API modules 710, 720, 730. In some embodiments, the plurality of domain modules 510, 520, 530, and 540 may include a first domain module 510, a second domain module 520, a third domain module 530, and a fourth domain module 540 The plurality of plug-in modules 610, 620 and 630 include a first plug-in module 610, a second plug-in module 620 and a third plug-in module 630, and the plurality of API modules 710 720 and 730 may include a first API module 710, a second API module 720 and a third API module 73, but the number of modules is not limited thereto.

The term "module" as used herein means a QA module. It is a module that selects and combines an unspecified number of pieces of information to form a knowledge or knowledge that can be utilized as a suitable solution for a given time and space. And a knowledge base such as a domain module, a mathematical operation, a unit conversion, and the like, which can independently perform a QA (query response) by referring to a knowledge base in which knowledge data is stored, A plug-in module that can be selectively loaded in the automatic question and answer system 1, and an API (application programming interface) that can be connected to an external service capable of providing QA, Module. Each QA module may be either a domain module, a plug-in module, or an API module. For example, the QA module includes a QA module such as a knowledge base QA module, a domain module, a mathematical operation plugin, a unit conversion plugin, a plugin module such as a time date plugin, a geography weather service, a currency conversion service, It can be an API module that can be connected with exchange rate QA module, DB based QA module, search based QA module and the like.

That is, the domain module, the plug-in module, and the API module are configured so that they can independently perform the QA in the automatic question-and-answer system 1. In this specification, for convenience of explanation, a domain module server 500 for managing a plurality of domain modules 510, 520, 530 and 540, a plug-in module server for managing a plurality of plug-in modules 610, 620 and 630, And the external service module server 700 that manages a plurality of API modules 710, 720, and 730 is illustrated as an example of the external service module server 600. However, the present invention is not limited thereto and a plurality of domain modules 510, 520, and 530 540, 540, a plurality of plug-in modules 610, 620, 630 and a plurality of API modules 710, 720, 730, or a plurality of domain modules 510, 520, 530, 540 Two or more module servers for managing a plurality of different modules of the plurality of plug-in modules 610, 620, and 630 and the plurality of API modules 710, 720, and 730 may be provided.

In this specification, the term "response" refers to an answer provided by the automatic question and answer system 1 for the query of the user 10, such as a query for the query of the user 10, The query result of the analysis result provided by each of the plurality of plug-in modules 610, 620 and 630 and the plurality of API modules 710, 720 and 730, Each of the plurality of domain modules 510, 520, 530, 540, the plurality of plug-in modules 610, 620, 630 and the plurality of API modules 710, 720, 730 may mean an answer. In order to distinguish from the "answer" of the automatic question and answer system 1 which is the correct answer to the inquiry of the user 10, a plurality of domain modules 510, 520, 530, 540, a plurality of plugin modules 610, 620 , 630 and each of the plurality of API modules 710, 720, 730 may be referred to as "answers ".

The automatic question and answer system 1 further includes a natural language generation unit 400 so that the question and answer manager 200 can provide the user 10 with a response in a natural language form.

That is, the automatic question-answering system 1 analyzes the query of the user 10 in a natural language form by the query analysis unit 210 with reference to the natural language understanding unit 300, 620, and 630 and a plurality of API modules 710, 720, and 730 and analyzes the analysis results in the modules of the natural language generation unit 400 And can generate the response in the form of a natural language in the Q & A manager 200 and provide it to the user 10.

The query response manager 200 includes a query analysis unit 210, a module selection execution unit 220, and a query response result integration unit 230. The query analyzing unit 210 can perform a natural language analysis on the query of the user 10 by referring to the natural language understanding unit 300.

For example, the natural language understanding unit 300 performs a plurality of unit analysis processes such as Semantic Role Labeling, morphological analysis, syntax analysis, entity name analysis, filtering analysis, intention classification, and domain analysis Can be performed. Some of the plurality of unit analysis processes may be performed in parallel. Some of the plurality of unit analysis processes may be performed sequentially.

In some embodiments, the natural language understanding unit 300 may provide the natural language analysis result of the query as a table structure or a structured data having a tree structure. The formatted data that can be provided by the natural language understanding unit 300 may be, for example, Relational DateBase (RDB), Comma-Seperated Variables (CSV), Extensible Markup Language (XML), JavaScript Object Notation (JSON) But is not limited thereto.

The query analyzing unit 210 analyzes at least a part of a plurality of unit analysis processes such as semantic role labeling, morphological analysis, syntax analysis, object name analysis, filtering analysis, intention classification, and domain analysis performed in the natural language understanding unit 300 So that the result of the natural language analysis on the query can be generated. In addition, the query analyzer 210 may transmit the natural language analysis result of the query to the module selection execution unit 220.

The module selection execution unit 220 selects some of the plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, 730 by referring to the natural language analysis result of the query, To be performed. In other words, the module selection execution unit 220 determines whether or not the automatic question and answer system 1 has a plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, The QA can be performed only on a few selected modules by referring to the natural language analysis result.

Therefore, QA can be performed only by a suitable one of the plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, 730. Thus, the load of the automatic question answering system 1 is minimized, The time required to obtain a response to the query of the server 10 can be shortened.

The module selection execution unit 220 receives a QA result performed in a selected one of a plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, 730, To the integration unit 230.

The query response result integrating unit 230 may integrate the QA results of the selected modules and provide a response to the user 10 through the user interface 100. [ The query response result integrating unit 230 can provide a response to the user 10 in a natural language form by referring to the natural language generating unit 400. [

The module selection execution unit 220 may select only a few modules suitable for the query of the user 10 and perform the QA. Therefore, in some embodiments, the query response result integrating unit 230 may generate the natural language word referring to the natural language generating unit 400 as a result of the QA performed in some selected modules, 0.0 > 100). ≪ / RTI > In some embodiments, the query response result integrating unit 230 determines a ranking for the QA result performed in the selected modules, selects a result having a high ranking, and transmits the result to the user 10 through the user interface 100 Can provide a response. In some embodiments, when two or more results of high similarity among the QA results performed on selected modules are received, the query response result integrating unit 230 selects the corresponding answer as a response and transmits the response to the user interface 100 To the user (10).

The automatic question answering system 1 according to the embodiment of the present invention can provide a correct response to the user 10 at the same time while minimizing the load and shortening the time in each of the QA execution and the response generation process.

2 is a block diagram illustrating an operation of a query analyzing unit of an automatic question-answering system according to an exemplary embodiment of the present invention.

Referring to FIG. 2, the query analyzing unit 210 can perform a natural language analysis on a query of a user (see 10 in FIG. 1) by referring to the natural language understanding unit 300. The query analyzing unit 210 may perform a natural language analysis on the query by referring to the natural language understanding unit 300 independently or may perform a natural language analysis on the natural language understanding unit 300 and then manage the result .

The natural language understanding unit 300 includes a semantic role labeling unit (SRL) 310, a morphological analysis unit 320, a syntax analysis unit 330, an entity name analysis unit 340, a filtering analysis unit 350, 360 and a domain analysis unit 370. The semantic roll labeling unit 310 may assign a semantic role to the query. The morpheme analysis unit 320 can separate the query assigned with the semantic role into morpheme units. The parser 330 and the entity name analyzer 340 can perform parsing and object name analysis on queries separated by morpheme units, respectively. The filtering analysis unit 350 may generate a simplified query pattern by removing unnecessary features included in the query. The intention classification unit 360 and the domain analysis unit 370 can perform intention classification and domain analysis of a query to which a meaning role is assigned based on the simplified query pattern generated by the filtering analysis unit 350. [

In some embodiments, the natural language understanding unit 300 may refer to the dictionary dictionary information 800 and assign a shoulder number to the morpheme separated by the morpheme analysis unit 320. [ Specifically, referring to the dictionary dictionary information 800, the intention classification unit 360 can assign a shoulder number to the separated morpheme to distinguish the meaning of the homonym.

The query analyzing unit 210 analyzes the natural language by referring to the natural language understanding unit 300. The natural language analyzing unit 210 analyzes the natural language by referring to the natural language analysis unit 300, And domain information, and provides the natural language analysis result to the module selection execution unit 220. In some embodiments, the intention information may be generated based on the descendant principle (5W1H, who, when, where, what, how, why) And the intention information may have shoulder number information.

3 is a block diagram illustrating the operation of the module selection execution unit of the automatic question answering system according to an exemplary embodiment of the present invention.

3, the module selection execution unit 220 may include a query response operation predicting unit 221, a query response module selection unit 222, and a query response module execution unit 223.

The query response operation predicting unit 221 receives the natural language analysis result from the query analyzing unit 210 and inputs the query response module operation prediction model 225 learned through the query response module selection model learning unit 224 and the manager The fitness of the query operation, which is a fitness for the queries of the modules, can be predicted based on the predetermined query response module operation rule 226. [ For example, in the query "1 + 1 =?", The QA operation predicting unit 221 can estimate the query operation fitness to 100% for the math QA module and 0% for the other QA modules. For example, the question-and-answer motion estimating unit 221 can predict the query operation fitness by the knowledge base-based QA module 80% and the search service 50%.

The question and answer module selection model learning unit 224 includes a question and answer module log 227 having feedback of the user 10 on question and answer results and / or question and answer results generated in each module, The learning set 228 can be used to learn the query response module motion prediction model 225. The query-and-response module selection model learning unit 224 may be implemented by, for example, a machine learning algorithm such as a linear regression, a logistic regression, or a regularization or a DNN (Deep Neural Network), a CNN (Convolutional Neural Network ), And RNN (Recurrent Neural Network (RNN)), for example, to learn the query response module operation prediction model 225.

The query response module selection unit 222 may refer to both the module characteristics stored in the query response module property storage unit 229 and the query operation suitability predicted by the query response operation predicting unit 221 to select a module for performing the QA have.

The query response module property storage unit 229 stores a plurality of modules (510, 520, 530, 540, 610, 620, 630, 710, 720, and 730 of FIG. 1) Module properties can be stored. The module property may be the Q & A coverage, which is the coverage of the QA that can be performed by the module. For example, the module characteristic may have information such as feature information, domain information, query pattern, number (number) information, ordinal information, alignment information, etc., QA performance or QA performance probability.

Therefore, the query response module selection unit 222 can select a module to perform the QA by referencing the query performance suitability of each query with respect to the query obtained through learning and the module characteristics as the query response coverage set for each module together.

The question and answer module execution unit 223 executes only the modules selected by the query response module selection unit 222 among the plurality of modules 510, 520, 530, 540, 610, 620, 630, 710, 720, The results of the analysis can be provided so that QA can be performed.

4 is a block diagram for explaining the operation of the query response operation predicting unit and the query response module selecting unit of the automatic question and answer system according to the exemplary embodiment of the present invention.

Referring to FIG. 4 together with FIG. 3, the query response operation predicting unit 221 can obtain a query operation fitness per module with respect to the query of the user 10 from the query response module operation prediction model 225. The query response module selection unit 222 can also select a module to perform the QA by referring to the query operation suitability for each module and the query response coverage for each module included in the query response module property storage unit 229. [

For example, if the query response module selector 222 is set to have a threshold of 60% for each of the query operation suitability and the query response coverage, the module with the query operation suitability and the query response coverage all being equal to or greater than the threshold value, If there are more than this schedule number, QA can be performed by selecting the relevant module. On the other hand, if there is no module having both the query operation suitability and the QA coverage threshold equal to or higher than the threshold value but the query operation suitability and the QA coverage are higher than the threshold value, the QA can be performed by selecting the corresponding module. Or a module in which both the query operation suitability and the query response coverage are both less than a certain number and the query operation suitability and the query response coverage are all equal to or more than the threshold value and one of the query operation suitability and the query response coverage is equal to or greater than the threshold value Modules can be selected together to perform QA.

At this time, the thresholds for the query operation suitability and the query response coverage may be set to the same value, but they may be set to be different from each other. The minimum number of modules to be selected by the QA module selection unit 222 to perform the QA is determined by considering the load of the automatic question and answer system (1 of FIG. 1) and the number of modules of the module servers 500, 600, and 700 , A positive integer (1, or 2 or more).

3, the QA results of the modules selected by the QA module selecting unit 222 are integrated in the QA unit 230 and transmitted to the user 10 through the user interface 100 . ≪ / RTI >

When the feedback of the response provided by the user 10 is inputted through the user interface 100, the feedback is stored in the query response module log 227, and the query response module selection model learning device 224 generates a query response module operation prediction model Lt; RTI ID = 0.0 > 225 < / RTI >

5 is a block diagram illustrating an automated query response system in accordance with an exemplary embodiment of the present invention.

5, the automatic question-and-answer system 2 includes a user interface 100 that receives a query from the user 10 via the network 20 and transmits a response, a query response manager 200, A natural language generating unit 400, module servers 500, 600, and 700, and a domain DB unit 550. [ In some embodiments, the automatic question and answer system 2 may further include a module model 900.

The query response manager 200 may include a query analysis unit 210, a module selection execution unit 220, and a query response result integration unit 230. The query analysis unit 210 analyzes the query of the user 10 in a natural language form with reference to the natural language understanding unit 300 and then analyzes the query in the domain module 510, 530, and 540, a plurality of plug-in modules 610, 620, and 630, and a plurality of API modules 710, 720, and 730 to perform QA, 400 in the form of a natural language response in the query response result integration unit 230 and provide the response to the user 10. [ When the user 10 inputs the feedback on the response provided by the user 10 through the user interface 100, the query response result integration unit 230 outputs the feedback of the user 10 to the module selection execution unit 220, .

The natural language understanding unit 300 includes a semantic role labeling unit (SRL) 310, a morphological analysis unit 320, a syntax analysis unit 330, an entity name analysis unit 340, a filtering analysis unit 350, 360 and a domain analysis unit 370. The semantic roll labeling unit 310 may assign a semantic role to the query. The morpheme analysis unit 320 can separate the query assigned with the semantic role into morpheme units. The parser 330 and the entity name analyzer 340 can perform parsing and object name analysis on queries separated by morpheme units, respectively. The filtering analysis unit 350 may generate a simplified query pattern by removing unnecessary features included in the query. The intention classification unit 360 and the domain analysis unit 370 can perform intention classification and domain analysis of a query to which a meaning role is assigned based on the simplified query pattern generated by the filtering analysis unit 350. [

The natural language generation unit 400 may include a natural language generation rule unit 410. The query response result integrating unit 230 can generate a natural language response by referring to the natural language creation rule unit 410. [

Module sub-servers 500, 600 and 700 may include at least one of a domain module server 500, a plug-in module server 600 and an external service module server 700, for example.

The first to fourth domain modules 510, 520, 530, and 540 of the domain module server 500 may include, for example, a Factoid Based Question Answering (FBQA) module, a Template Based Question Answering (TBQA) Based Question Answering module, a DNQA (Deep Neural Based Question Answering) module, or an IRQA (Information Retrieval Based Question Answering) module.

The first to fourth domain modules 510, 520, 530 and 540 of the domain module server 500 are connected to at least one of the plurality of domain databases 551, 552, 553 and 554 of the domain DB unit 550 By referring to the domain DB, an answer can be generated. The plurality of domain DBs 551, 552, 553, and 554 may be knowledge bases in which knowledge data is accumulated. The domain DBs 551, 552, 553 and 554 may include, for example, a common knowledge base, a history domain knowledge base, a person domain knowledge base, a movie domain knowledge base, a music domain knowledge base, a health domain knowledge base, Base, or food restaurant domain knowledge base.

The first to third plug-in modules 610, 620 and 630 of the plug-in module server 600 can be configured to perform the QA on their own without referring to the knowledge base and can be installed in the automatic question and answer system 2 have. The first to third plug-in modules 610, 620, 630 may be, for example, a mathematical operation plug-in, a unit conversion plug-in, a time date plug-in, a distance calculation plug-in, or a currency conversion plug-

The automatic question and answer system 2 may be connected to the external service DB 750 through the external service module server 700 and the network 20a. The network 20a may be of the same type as the network 20 in which the user 10 sends and receives query responses to the user interface 100, and thus a detailed description thereof will be omitted. The first to third API modules 710, 720 and 730 of the external service module server 700 may have APIs for connecting external services 751, 753 and 755 capable of performing QA. For example, the external service module server 700 may include a first API module 710, a second API module 720, and a third API module 730, and the first API module 710, The second API module 720 and the third API module 730 may be connected to the first service 751, the second service 753 and the third service 755 via the network 20a. The first service 751, the second service 753 and the third service 755 are connected to a first external DB 752, a second external DB 754 and a third external DB 756, respectively, Or may include. For example, the first service 751, the second service 753, and the third service 755 may be a geography weather service, a currency conversion service, or a search service, and may include a first external DB 752, 2 external DB 754 and the third external DB 756 may be a geographical weather DB, an exchange rate information DB, or a search DB.

Module model 900 may include query response logic 910, query response execution conditions 920, and natural language generation rules (NLG Rule 930). The query response logic 910 provides logic for generating an answer from the results of unit natural language analysis for a query in a plurality of domain modules 510, 520, 530, 540 and a plurality of plugin modules 610, 620, 630 can do. The query response execution condition 920 is used to generate a response from a unit natural language analysis result of a query in a plurality of domain modules 510, 520, 530, 540 and a plurality of plugin modules 610, 620, 630 You can provide the conditions you need. The natural language generation rule 930 refers to the natural language generation rule part 410 from the QA results performed by the plurality of domain modules 510, 520, 530 and 540 and the plurality of plugin modules 610, 620 and 630, You can provide a rule to create a form of answer.

6 is a detailed block diagram illustrating a domain module server and a domain DB unit of the automatic question and answer system according to an exemplary embodiment of the present invention.

6, the domain module server 500 includes a domain module 580 having a plurality of domain modules 510a, 520a, 530a, and 540a, and a plurality of domain modules 510a , 520a, 530a, and 540a, respectively. The domain module registration unit 590 may have a query response coverage for each of a plurality of domain modules 510a, 520a, 530a, and 540a.

The plurality of domain modules 510a, 520a, 530a, and 540a may be FBQA module 510a, TBQA module 520a, GBQA module 530a, and DNQA module 540a, respectively. The FBQA module 510a can automatically generate a SPARQL-type query through keyword-based pattern matching of a short-form query, and then generate an answer by querying the domain DB unit 550, i.e., the knowledge base. The TBQA module 520a may generate a response by querying the domain DB unit 550, that is, the knowledge base, after completing the template-based query by performing template mapping from the query based on previously prepared templates. The GBQA module 530a analyzes a query of a more complex form than a short sentence by using a syntax analysis or the like to automatically generate a SPARQL type query and generate an answer by querying the domain DB unit 550, have. The DNQA module 540a may generate a query using a Deep Learning Neural Network and may generate an answer by querying the domain DB portion 550, i.e., the knowledge base.

In some embodiments, the domain module portion 580 may further include an IRQA module. The IRAQ module can generate a search-based answer to a query.

The domain DB unit 550 includes a common domain knowledge base 551a, a history domain knowledge base 552a, a person domain knowledge base 553a, and a movie domain knowledge base 554a. Knowledge base. In some embodiments, the domain DB portion 550 may further include a knowledge base such as a music domain knowledge base, a health domain knowledge base, and the like.

Specifically, the FBQA module 510a can process a short-form query. For example, a query such as "height of Mt. Halla ?," It is possible to find a class such as an entity and a property and generate a response to a query through a predefined SPARQL corresponding to a pattern EP having the corresponding classes. For example, in the natural language generation rule (930 in FIG. 5), when the rule provided for the corresponding EP pattern is a short response type, an answer of 1950m is generated. The FBQA module 510a includes an EP (Entity, Property) pattern, an EPP (Entity, 2 Property) pattern (for example, a property: , And the EPQ (Entity, Property, Quntity) pattern (for example, how many members of the IOI (class: Artist) are the quantity?). .

 Specifically, the TBQA module 520a performs a more complex type of QA than a short sentence. It generates an SPARQL by classifying the intention of a query according to a domain, mapping the intention to a SPARQL template defined in each intention, You can create an answer to the query. That is, the TBA module 520a may process queries specific to a particular domain. For example, a query such as " What is the main character of the girl disappearing with the wind? "Is mapped to a SPARQL template defined in advance in the intention of the query to generate a SPARQL for the query, To generate an answer.

For example, the TBQA module 520a may determine whether the intention of the query is related to the movie content, the movie actor, the movie screening (theater, the time, etc.) Of SPARQL templates, depending on the intent of the detailed query, you can have more SPARQL templates.

Therefore, the plurality of domain modules 510a, 520a, 530a, and 540a of the domain module unit 580 may have different query operation suitability according to the analysis result of the query of the natural language understanding unit 300 (FIG. 5).

7 is a block diagram illustrating the plug-in module server of the automatic question-answering system according to an exemplary embodiment of the present invention in detail.

7, the plug-in module server 600 includes a plug-in module 680 having a plurality of plug-ins 610a, 620a, and 630a, and a plurality of plug-ins 610a, 620a, And a plug-in register 690 having information on each of the plug-in registers 630a and 630a. The plugin register 690 may have a query response coverage for each of the plurality of plugins 610a, 620a, 630a.

The plurality of plugins 610a, 620a, and 630a may be a mathematical operation plug-in 610a, a unit conversion plug-in 620a, and a time date plug-in 630a, respectively.

The mathematical operation plug-in 610a can itself generate an answer to a query regarding mathematical operations such as arithmetic operations. The unit conversion plug-in 620a can itself generate an answer to the inquiry about the unit conversion. The time date plug-in 650a can itself generate responses to queries such as time, date queries, world clocks, volume transitions, and the like.

Therefore, the plurality of plug-ins 610a, 620a, and 630a included in the plug-in module unit 680 can generate the answers themselves without referring to the analysis result of the query of the natural language understanding unit 300 The quality of the query operation may be different.

8 is a block diagram illustrating in detail an external service module sub and an external service / DB coupled thereto according to an exemplary embodiment of the present invention.

8, the external service module server 700 includes an API module 780 having a plurality of API modules 710a, 720a, and 730a, and a plurality of API modules 710a, And an external service module registration unit 790 having information on each of the external service modules 720a and 730a. The external service module registration unit 790 may have a query response coverage for each of a plurality of external services 751a, 753a, and 755a that are connected to the plurality of API modules 710a, 720a, and 730a, respectively, .

The plurality of API modules 710a, 720a, and 730a may be connected to the external service / DB 750 through the network 20a.

The plurality of API modules 710a, 720a, and 730a may be, but not limited to, a geography weather service API 710a, a currency conversion service API 720a, and a search service API 730a.

The external service / DB 750 may include, for example, a geography weather service 751a and a geographical weather DB 752a, a currency conversion service 753a associated with or included therein, and a currency exchange information DB 754a ), And search service 755a, and a search DB 756a associated with or included therein, but are not limited thereto.

The geography weather service API 710a may have an API to be associated with the geography weather service 751a and may be provided to the geography weather service 751a via a geographic weather service 751a to determine the distance or path between the two regions, You can generate answers to queries such as weather. The exchange conversion service API 720a may have an API to be associated with the exchange rate conversion service 753a and may generate an answer to the query such as the current exchange rate, exchange rate between different currencies, etc. through the exchange rate conversion service 753a . The search service API 730a may have an API to be connected to the search service 755a and may generate an answer to the query through the search service 755a.

Therefore, the plurality of API modules 710a, 720a, and 730a of the API module unit 780 can obtain an answer to itself by referring to the external service / DB as a result of the analysis of the query of the natural language understanding unit 300 The quality of the query operation may be different.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, This is possible.

The present invention relates to an automatic query response system and a method and system for automatically inquiring response to a questionnaire response message in response to a request from a user, Domain DB unit 600: plug-in module server 700: external service module server 900: module / plug-in model

Claims (10)

A user interface for receiving a query in natural language form from a user via a network and sending a response;
At least one module server having a plurality of QA modules capable of performing a query response;
A query analyzer for performing a natural language analysis on the query;
A module selection execution unit for selecting some of the plurality of QA modules based on a natural language analysis result of the query; And
And a query response result integration unit for providing the response to the query to the user via the user interface,
Wherein the module selection execution unit comprises:
A query response operation predicting unit that receives a natural language analysis result of the query and predicts a query operation fitness for each of the plurality of QA modules for the query through learning;
A query response module selection unit for selecting some of the plurality of QA modules with reference to the query response coverage set for each of the plurality of QA modules and the query operation suitability; And
And a query response module executing unit for providing a natural language analysis result of the query to some of the plurality of QA modules selected by the query response module selection unit to perform a query response. The method according to claim 1,
Wherein the module selection execution unit further comprises a query response module selection model learning unit that learns a query response module operation prediction model,
Wherein the query response operation predicting unit predicts the query operation fitness for each of the plurality of QA modules based on the learned query response module operation prediction model and the query response module operation rule set in advance. 3. The method of claim 2,
Wherein the query response module selection model learning device comprises:
Wherein the learning model learning unit learns the query response module operation prediction model using the query response result performed in each of the plurality of QA modules and the feedback on the response of the user. The method according to claim 1,
Further comprising a plurality of domain DBs each of which is a knowledge base on which knowledge data is accumulated,
Wherein some of the plurality of QA modules generate an answer with reference to at least one of the plurality of domain DBs. 5. The method of claim 4,
Wherein the plurality of QA modules are plug-in modules that generate their own responses without referring to the plurality of domain DBs. 5. The method of claim 4,
Wherein some of the plurality of QA modules are connected to an external service through a network to generate an answer. 5. The method of claim 4,
Some of the modules referring to at least one of the plurality of domain DBs among the plurality of QA modules include an FBQA module for generating a SPARQL type query through pattern matching of a keyword based on the user's query to generate a response, A TBQA module for generating a query based on a template by performing template mapping from the user's query based on the templates, a TBQA module for analyzing the user's query by parsing to generate a SPARQL-type query, Module, and a DNQA module for generating a query using a Deep Learning Neural Network to generate an answer. The method according to claim 1,
Further comprising a natural language understanding unit referenced by said query analysis unit,
Wherein the natural language understanding unit performs semantic role labeling, morphological analysis, syntax analysis, object name analysis, filtering analysis, intent classification, and domain analysis. 9. The method of claim 8,
Wherein the natural language analysis result of the query includes intention information based on the intention classification,
Wherein the intention information is assigned with reference to language dictionary information in the morpheme separated through the morphological analysis and has a shoulder number for distinguishing the meaning of the homonym. The method according to claim 1,
Wherein the query response module selection unit comprises:
Wherein at least one of the query response coverage and the query operation fitness among the plurality of QA modules is selected to be equal to or larger than a set threshold value.

Images