KR101849750B1 - Interactive Automatic Question-Answering System - Google Patents

Abstract

The present invention provides an interactive automatic question-answering system which can accurately understand a question of a user, and quickly and accurately answer the question. According to an aspect of the present invention, the interactive automatic question-answering system comprises: a dialog manager to receive a question of a user in a natural language form via a network, transmit an answer, and interactively give and take questions and answers from the user; a question-answering manager to manage the question of the user and a response to the question; a natural language understanding unit having a plurality of unit analysis processes to analyze the question of the user in a natural language form and generate unit natural language analysis results; and a plurality of modules and a plurality plug-ins to receive the unit natural language analysis results generated during the plurality of unit analysis processes of the natural language understanding unit to independently perform analysis. The question-answering manager classifies the unit natural language analysis results generated during the plurality of unit analysis processes of the natural language understanding unit to receive the unit natural language analysis results to transfer the unit natural language analysis results to a module or a plug-in corresponding to each unit natural language analysis result among the plurality of modules and the plurality of plug-ins to allow the module and the plug-in to perform analysis, and allows the user to ask a question by the dialog manager if an answer is not received from the plurality of modules and the plurality of plug-ins or additional information is requested.

Description

[0001] Interactive Automatic Question-Answering System [

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an interactive automatic question-and-answer system, and more particularly to an interactive automatic question-answering system based on a QA (Question-Answering) module and a QA plug-in.

The present invention is derived from research conducted and conducted by Saltlux Co., Ltd. as part of the SW Technology Computing Industry Source Technology Development Project (SW) of the future Creation Science Department. [Research period: 2006.03.01 ~ 2017.02.28, Research institute: Information and Communication Technology Promotion Center, Research title: WiseKB: Big data understanding based self-learning knowledge base and reasoning technology development, task number: R0101- 16-0054]

As the amount of data on the Web exponentially grows, today's Web is reaching the point where almost all of the information you want is available. Therefore, in order to provide the information desired by the user, it is necessary to appropriately extract the data existing on the web according to the user's demand. Accordingly, information search researches for searching documents related to the user's query are actively conducted, An information retrieval system was created.

However, the conventional web-based information retrieval system is limited in that it provides a related document when a user requests a direct answer. Therefore, there is a need for a query and response system that takes a step further in an information retrieval system that simply provides a related document, grasps a user's query clearly, and responds to the answer itself.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an interactive automatic question and answer system capable of accurately understanding a user's query and promptly and accurately responding to a query.

According to an aspect of the present invention, there is provided an interactive automatic question-answering system for receiving a query of a user in a natural language form through a network and transmitting a response, A query manager responsive to the user's query and the response to the query, a plurality of unit analysis processes for analyzing a query of the user in a natural language form and generating a unit natural language analysis result, And a plurality of modules and a plurality of plugins that receive the unit natural language analysis results respectively generated in the plurality of unit analysis processes of the natural language understanding unit and perform independent analysis respectively, And the unit natural < RTI ID = 0.0 > Analyzing the result of analysis and delivering the analyzed result to a module or a plug-in corresponding to each of the plurality of modules and the plurality of plug-ins, and receiving an answer from the plurality of modules and a plurality of plug- Or to ask the user through the dialog manager if additional information is requested.

The dialog manager may combine the user's answer to the query with the user's query and send it to the query response manager.

Wherein the dialog manager comprises a user memory for storing a query of the user for the current query, a short memory storing the question and the user's answer, and a previous query with the user, a question and an answer thereto, The question may be generated by referring to the previous query with the user stored in the user memory, a question and an answer thereto.

The query response manager may generate the response on the basis of a response transmitted to a predetermined number of times or a predetermined number of times in a module or plug-in under analysis among the plurality of modules and a plurality of plug-ins.

When the predetermined time period has elapsed, the query response manager may stop analysis of a module or plug-in that has not transmitted a response, out of modules or plug-ins being analyzed.

The query response manager may stop analysis of a module or plug-in that has not delivered a response, out of the module or plug-in being analyzed, when a predetermined number of responses have been transmitted.

Wherein each of the plurality of modules generates an answer with reference to at least one of the plurality of domain DBs, and the plurality of plug-ins include a plurality of domain DBs You can generate an answer without referring to it.

Some of the plurality of plug-ins may generate responses themselves, and others may be associated with external services to generate responses.

Wherein the natural language understanding unit includes a semantic roll labeling unit, a morphological analysis unit, a syntax analysis unit, a object name analysis unit, an intention classification unit, and a domain analysis unit, each of which performs the plurality of unit analysis processes, The unit natural language analysis result generated in the unit analysis process performed in each of the morphological analysis unit, the syntax analysis unit, the object name analysis unit, the intention classification unit, and the domain analysis unit is transmitted to the query response manager Lt; / RTI >

Wherein the query response manager determines a similarity of answers transmitted in a predetermined time in a module or a plug-in under analysis among the plurality of modules and a plurality of plug-ins, and if the similarity of at least two answers is equal to or greater than a certain standard, Wherein the response is generated based on at least two answers that are equal to or greater than a certain criterion, and when at least two answers whose similarity is equal to or higher than the predetermined criterion are delivered, the module or plug- Can be stopped.

The interactive automatic question and answer system according to the present invention can interactively exchange questions and answers with a user, and can accurately understand a user's query, thereby enabling a quick and accurate question and answer.

And can only perform analysis on the required modules and / or plug-ins so that the load required to obtain a response to the query can be minimized and the analysis in each module and / or plug-in can be started quickly, Time can be shortened.

Also, if an answer to obtain a response from among a plurality of modules and a plurality of plugins is provided within a predetermined time, or if there is a certain number of times to be provided, the analysis can be stopped for other modules in process and / The load required to obtain the response can be minimized, and the time for obtaining the response can be shortened.

1 is a block diagram illustrating an interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention.
2 is a block diagram illustrating an interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention.
3 is a block diagram illustrating an operation of an interactive automatic question-answering system according to an exemplary embodiment of the present invention.
4 is a block diagram illustrating an operation of an interactive automatic question and answering system according to an exemplary embodiment of the present invention.
5 is a block diagram illustrating the operation of the interactive automatic question-answering system according to an exemplary embodiment of the present invention.
6 is a block diagram illustrating an interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention.
7 is a block diagram illustrating a module server and a domain DB unit of the interactive 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 / DB connected to a plug-in server and a plug-in server of an interactive automatic question-answering system 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 interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention.

Referring to Fig. 1, an interactive automatic question and answer system 1 includes a dialog manager 100 that receives a query from a user 10 via a network 20 and transmits 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 smart phone or a tablet, the network 20 may be a wireless data communication service such as 3G, long term evolution (LTE), wireless LAN such as Wi-Fi, Bluetooth,

The conversation manager 100 may provide an interface for accessing the interactive 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 interactive automatic question answering system 1 via the dialogue manager 100 and to the interactive automatic question answering system 1 via the dialogue manager 100 A response can be received.

The dialogue manager 100 can also interactively ask the user 10 to correctly understand the query in the event that it is difficult to give a clear answer to the query of the user 10, You can store and manage your answers.

The interactive automatic question and answer system 1 is connected to the conversation manager 100 to receive a query from the user 10 and to determine a response and to transmit a response to the user 10 via the conversation manager 100, And a question and answer manager 200 that allows the administrator 100 to exchange questions and answers. And can also control the components included in the interactive automatic question and answer system 1 by the question and answer management 200.

The interactive automatic question and answer system 1 includes a natural language understanding unit 300 for analyzing a natural language type query received by the query response manager 200 and a natural language analysis result of a query analyzed by the natural language understanding unit 300, A module server 500 that provides answers to be provided via the response manager 200 and provides a response to the query, and a plug-in server 600.

In the present specification, the term "response" means a correct answer to the question provided by the interactive automatic question and answer system 1 about the query of the user 1, for example, the query of the user 1, 620, 630, 640, 650, and 660 of the modules 510, 520, 530 and / or the plug-in server 600 of the module sub 500, 530 and / or each of the plug-ins 610, 620, 630, 640, 650, and 660 with respect to the query of the user 10. Quot; means that the dialog manager 100 makes a request to the user 10 to obtain additional information about the query, and the "answer" Quot; additional query ", which is the provision of additional information of the user 10 to the "query ". That is, the query that the interactive automatic question and answer system 1 can provide the "response " may include both the initial query and the answer (additional query) of the user 10. [

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 module capable of independently performing QA by referring to a knowledge base in which knowledge data, which is data having a relationship therebetween, is accumulated. Also, the term "plug-in " means a QA plug-in. The " plug-in " means that the QA can be performed by itself without referring to a knowledge base (domain DB) Those that can be selectively loaded, or have an application programming interface (API) that can connect with external services that can provide QA. That is, the modules and plug-ins are configured to be able to perform the QA within the interactive automatic question and answer system 1, independently of each other module or plug-in.

The interactive automatic question and answer system 1 further includes a natural language generation unit 400 so that each module 510, 520 and 530 and / or each of the plugins 610, 620, 630, 640, 650, Type answers, or the query response manager 200 may provide a natural-language response.

In other words, the interactive automatic question and answer system 1 analyzes the query of the user 10 in the natural language form in the natural language understanding unit 300, and then analyzes the query in the natural language type understanding unit 300 using the module server 500 and / And generates a response in the form of a natural language through the natural language generation unit 400 and provides the response to the user 10.

The query response manager 200 includes a notification receiving unit 210, a module / plug-in detecting unit 220, and a response determining unit 230. The notification receiving unit 210 can receive the result of the natural language analysis of the analyzed query in the natural language understanding unit 300. In some embodiments, the natural language understanding unit 300 does not transmit the analysis result to the notification receiving unit 210 after all the analysis is completed, To the notification receiving unit 210 sequentially in the order in which they are generated.

For example, the natural language understanding unit 300 performs a plurality of unit analysis processes such as Semantic Role Labeling, morphological analysis, syntax analysis, object name analysis, intent classification, and domain analysis on a query . 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 data structure or a structured data having a tree structure. The formatted data provided by the natural language understanding unit 300 may be, for example, Relational DateBase (RDB), Comma-Seperated Variables (CSV), XML (extensible Markup Language), or JSON (JavaScript Object Notation).

The notification receiving unit 210 receives at least one unit analysis process among a plurality of unit analysis processes such as semantic role labeling, morphological analysis, syntax analysis, object name analysis, intent classification, and domain analysis performed in the natural language understanding unit 300 The unit natural language analysis result can be separately received. The notification receiving unit 210 may store the unit natural language analysis results received by the natural language understanding unit 300 until the response to the query is completed. In addition, the notification receiving unit 210 may transmit the unit natural language analysis results received by the natural language understanding unit 300 to the module / plug-in detection unit 220. [

The module / plug-in detection unit 220 transmits the unit natural language analysis result received from the notification receiving unit 210 to the module sub 500 and / or the plug-in server 600 based on the unit natural language analysis result received from the natural language understanding unit 300 620, 630, 640, 650, and 660 of the modules 510, 520, and 530 of the module sub 500 and / or plug-in servers 610, 620, 630, 640, 650, and 660 of the plug- Allowing the appropriate modules and / or plug-ins to analyze the answer to the query.

Accordingly, the module / plug-in detection unit 220 can detect the presence of a plurality of modules 510, 520, 530 and a plurality of plug-ins 610, 620, 630, 640, 650, 660 of the interactive automatic question- And to perform analysis only for the appropriate modules and / or plug-ins. Further, even before the completion of the analysis in the natural language understanding unit 300 for the query, a unit natural language analysis result as a partial analysis result is received, and an appropriate module and / or plugin corresponding to the unit natural language analysis result is first Can be driven. Therefore, it is possible to perform only analysis for necessary modules and / or plug-ins, so that the load required to obtain a response in the interactive automatic question and answer system 1 is minimized, and the load on modules 510, 520, 530 and / 610, 620, 630, 640, 650, 660) can be started quickly and the time for obtaining a response can be shortened.

A module and / or plug-in that has performed an analysis of the query among a plurality of modules 510, 520, 530 and a plurality of plug-ins 610, 620, 630, 640, 650, 660, And transmits it to the determination unit 230. The response determination unit 230 may determine the response by itself or by referring to the answer ranking unit 240 and then provide the response to the user 10 through the dialogue manager 100. [ Also, the response determining unit 230 may determine whether additional information is required to respond to the query, and may cause the conversation manager 100 to ask the user 10 to request an additional query. The response determining unit 230 can refer to the natural language generating unit 400 and provide a response or a question to the user 10 in a natural language form.

In some embodiments, when modules 510, 520, 530 and / or plug-ins 610, 620, 630, 640, 650, 660 perform analysis and deliver responses to response determiner 230, And transmits the answer in the natural language form to the response determining unit 230 with reference to the response determining unit 400. [

In some embodiments, the response determination unit 230 may determine only responses that are delivered within a certain time out of the plurality of modules 510, 520, 530 and the plurality of plug-ins 610, 620, 630, 640, 650, 660 After receiving the answer, the user can determine the ranking by referring to the answer ranking unit 240, and select a response having the highest ranking among the answers as a response, and provide the answer to the user 10. In this case, if the predetermined time has elapsed, the control unit may be configured to notify the module and / or the plug-in which analysis is in progress among the plurality of modules 510, 520, and 530 and the plurality of plug-ins 610, 620, 630, 640, 650, and 660 The analysis can be stopped.

In some embodiments, the response determination unit 230 may determine that two or more responses having high similarity among the plurality of modules 510, 520, 530 and the plurality of plug-ins 610, 620, 630, 640, 650, It is possible to select the answer as a response and provide it to the user 10. In this case, when two or more answers having high similarity are received, a plurality of modules 510, 520, 530 and a plurality of plug-ins 610, 620, 630, 640, 650, / / Stop the analysis on the plugin.

In some embodiments, the response determination unit 230 may not receive a response from the plurality of modules 510, 520, 530 and the plurality of plugins 610, 620, 630, 640, 650, If additional information is requested, the user can be prompted to ask the user 10 through the dialogue manager 100.

Therefore, by the operation of the response determining unit 230, the load required to obtain the response in the question answering system 1 is minimized, and the time for obtaining the response can be shortened.

The interactive automatic question answering system 1 according to an embodiment of the present invention minimizes the load and shortens the time to obtain a response in each of the query analysis process and the response decision process, .

2 is a block diagram illustrating an interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention. FIG. 2 is a detailed view of the response determining unit 230, and the description overlapping with the description of FIG. 1 may be omitted.

2, the interactive automatic question answering system 1a may include a conversation manager 100, a query response manager 200, a module server 500, and a plug-in server 600. [ The interactive automatic question and answer system 1a may further include a natural language understanding unit 300. [

When the interactive automatic question and answer system 1a receives a natural language query from the user 10 via the network 20, the natural language understanding unit 300 receives the query through the query response manager 200, And a plurality of unit analysis processes for natural language understanding. When a unit natural language analysis result for each unit analysis process among a plurality of unit analysis processes for a query being performed by the natural language understanding unit 300 is generated, the unit natural language analysis result is transmitted to the query response manager 200.

The module / plug-in detection unit 220 receives each of the received natural language analysis results from the plurality of modules 510 and 520 of the module sub 500 620, 630, 640, 650, and 660 of the plug-in server 600 and the appropriate module and / or plug-in corresponding to the result of the unit natural language analysis, The module and / or the plug-in receives the response, and transmits the response to the response determining unit 230.

The response determining unit 230 may include an additional query determining unit 232, an answer time managing unit 234, an answer similarity determining unit 236, and an answer ranking determining unit 238.

The answer time management unit 234 can recognize and manage the time and / or the number of times the answers are delivered from the plurality of modules 510, 520, 530 and the plurality of plugins 610, 620, 630, 640, 650, 660 have.

In some embodiments, the response time management unit 234 allows the responses to be delivered from the module and / or the plug-in to the response determining unit 230 only up to a predetermined time, and if a predetermined time has elapsed, You can stop the analysis at the plug-in, or stop it at one hour. Then, the answer ranking determining unit 238 refers to the answer ranking unit 240, determines a ranking for the answers delivered within a predetermined time, and selects a response with a high ranking among the answers as a response to the inquiry .

In some embodiments, the response time management unit 234 allows the responses to be forwarded from the module and / or plug-in to the response determination unit 230 until a certain number of times, and when a certain number of responses are delivered, The analysis in the module and / or plug-in can be interrupted or stopped at one hour. Then, the answer ranking determining unit 238 refers to the answer ranking unit 240, determines a ranking for a certain number of the transmitted answers, and selects an answer with a high ranking among the answers as a response to the inquiry .

In some embodiments, the answer time management unit 234 does not answer the response from the module and / or the plug-in to the response determining unit 230 until a predetermined time, does not forward a predetermined number of answers, It may transmit the response to the additional query determiner 232. [

The additional query determiner 232 may provide an additional query to the user 10 via the conversation manager 100 when the response is not transmitted from the response determining unit 230 or a response is received that the information for generating the answer is insufficient You can ask to send a question to ask. Or in some embodiments, the additional query determiner 232 may request the response determiner 230 to wait further for an answer from the module and / or plugin for a certain additional time.

The response similarity degree determiner 236 can determine the similarity of the responses transmitted from the plurality of modules 510, 520, and 530 and the plurality of plugins 610, 620, 630, 640, 650, and 660. For example, the response similarity determining unit 236 may determine that the similarity of at least two of the answers sent from the module and / or the plug-in to the response determining unit 230 is higher than a predetermined reference, You can stop the analysis at the plug-in, or stop it at one hour.

In some embodiments, if the similarity of at least two answers has a degree of similarity that is equal to or greater than a certain criterion, one of them, or a combination thereof, may be selected in response to the query.

In some embodiments, if the similarity of at least two answers has a degree of similarity that is equal to or higher than a predetermined standard, the answer ranking unit 238 refers to the answer ranking unit 240 for answers having a degree of similarity higher than a certain standard, After selecting the answer, you can select the answer with the highest ranking among the answers in response to the inquiry.

3 is a block diagram illustrating an operation of an interactive automatic question-answering system according to an exemplary embodiment of the present invention. The description overlapping with the description of FIG. 1 or FIG. 2 in the description of FIG. 3 may be omitted.

3, the interactive automatic question answering system 1b may include a dialog manager 100, a query response manager 200, a module server 500, and a plug-in server 600. [

The response determination unit 230 may include an additional query determination unit 232 and an answer time management unit 234. [ When the information for generating an answer is insufficient in the course of performing the analysis, the plurality of modules 510, 520, 530 and the plug-ins 610, 620, 630, 640, 650, , Or may request additional information from the response determining unit 230. [

For example, if the response time management unit 234 does not receive a response from the plurality of modules 510, 520, 530 and the plug-ins 610, 620, 630, 640, 650, 660 until the first predetermined time , It may notify the additional query determiner 232 that the answer is not transmitted. In this case, the additional query determiner 232 can ask the user 10 via the conversation manager 100, such as " repeat the query ". When the user 10 answers to this, the conversation manager 100 combines the initial query of the user 10 and the answer to the query together and transmits the combined query to the query response manager 200, .

For example, when additional information is requested from a plurality of modules 510, 520, and 530 and plug-ins 610, 620, 630, 640, 650, and 660, the response time management unit 234 may include an additional query determiner 232 May be notified of the request for additional information. In this case, the additional query determiner 232 may ask the user 10 through the conversation manager 100 to request additional information. When the user 10 answers to this, the conversation manager 100 transmits the initial query of the user 10 and the answer to the question together to the query response manager 200 so that the analysis of the query is performed again . For example, when the user 10 makes a query of "What is the weather", the location and time information from the plurality of modules 510, 520, 530 and plugins 610, 620, 630, 640, 650, May request the response time management unit 234 to request the response time management unit 234. [ In this case, the additional query determiner 232 notifies the user 10 via the conversation manager 100 of a question such as "Which weather do you ask?" Or " can do. When the user 10 answers "Seoul" or "yes" to the question, the conversation manager 100 combines the existing query and the user's answer together and transmits to the question and answer manager 200, "Or" What is the current weather in Seoul? ".

Such a question and answer through the dialogue manager 100 can be repeatedly performed until an accurate answer to the inquiry is obtained, rather than ending one session.

For example, if the user 10 made an initial query of "How can I know if I am fat," a plurality of modules 510, 520, 530 and plugins 610, 620, 630, 640, 650, 660), you may be asked to require a BMI index. In this case, the conversation manager 100 may ask the user 10, such as "Would you like to see the BMI index?" If the user 10 answers yes to this, the dialogue manager 100 may send a query to the query response manager 200, "What is my BMI index?" A plurality of modules 510, 520, 530 and plug-ins 610, 620, 630, 640, 650, 660 can make a request that keys and weight are needed. In this case, the conversation manager 100 may ask the user 10 "What is the key and the weight" again, and when the user 10 answers the key and weight, the interactive automatic question and answer system (1b) can respond like "BMI index is XX" or "BMI index is XX, so it's normal".

In some embodiments, instead of having the dialog manager 100 ask a question, the additional query determiner 232 may provide a plurality of modules 510, 520, 530 and plug-ins 610, 620 , 630, 640, 650, 660). In this case, the answer time management unit 234 may wait for an additional time during an additional time based on the request to the additional query determination unit 232. [

4 is a block diagram illustrating an operation of an interactive automatic question and answering system according to an exemplary embodiment of the present invention. 4 is a block diagram for explaining an operation after the completion of a query and an answer between the user 10 and the conversation person 100 described in FIG. 3, and FIG. 4 And the description of FIG. 3 may be omitted.

4, the interactive automatic question answering system 1c may include a conversation manager 100, a query response manager 200, a module server 500, and a plug-in server 600. [ The interactive automatic question and answer system 1c may further include a natural language understanding unit 300. [

When the interactive automatic question and answer system 1c receives the natural language query from the user 10 via the network 20, the natural language understanding unit 300 receives the query through the query response manager 200, And a plurality of unit analysis processes for natural language understanding. When the notification receiving unit 210 receives the unit natural language analysis result of at least one unit analysis process among the plurality of unit analysis processes for the query being performed by the natural language understanding unit 300, the module / plug- 620, 630, 640, 650, and 660 of the plug-in server 600 and the plurality of modules 510, 520, and 530 of the module sub 500, To the appropriate modules and / or plug-ins corresponding to each.

For example, a first module 510, a second module 520, a first plug-in 610, and a second plug-in module 610 of a plurality of modules 510, 520, and 530 and plugins 610, 620, 630, 640, 650, And the second plug-in 620 receive the unit natural language analysis result, and can independently perform the analysis. At this time, the unit natural language analysis results received by the first module 510, the second module 520, the first plug-in 610, and the second plug-in 620 are transmitted to the natural language understanding unit 300, The result may be the same unit natural language analysis result generated in the process, or the unit natural language analysis result generated in the other unit analysis process. The third module 530 and the third through sixth plug-ins 630, 640, 650, and 660 that have not received the unit natural language analysis result are in an idle state without performing analysis, Can be performed.

The modules and / or plug-ins that have been analyzed and generated answers, among the first module 510, the second module 520, the first plug-in 610, and the second plug-in 620 under analysis, The response is sent to the response determination unit 230 without considering whether to perform the analysis of the module and / or the plug-in.

In some embodiments, the response time management unit 234 can manage the elapsed time from the point in time when the unit natural language analysis result is first transmitted from the module / plug-in detection unit 220 to the module server 500 and / or the plug-in server 600 have. In some embodiments, the answer time management unit 234 can manage the elapsed time from the point in time when the answer is first delivered to the response determination unit 230. [

The response time management unit 234 may receive the answers until the elapsed time becomes the second predetermined time, and the response determination unit 230 may determine a response from the responses. In addition, the response determining unit 230 may stop the analysis in the module and / or the plug-in that has not yet generated the answer, or may stop the analysis in one hour, after the second time has elapsed.

The second time may be set by the administrator of the interactive automatic question answering system 1c or may be set by the interactive automatic question answering system 1c based on the time taken to respond to the previously performed query It can be set by itself.

The answer time management unit 234 may extend the elapsed time so that the answer is delivered to the response determination unit 230 during a third time greater than the second time when no answer is delivered during the second time.

5 is a block diagram illustrating the operation of the interactive automatic question-answering system according to an exemplary embodiment of the present invention. Specifically, FIG. 5 is a block diagram for explaining an operation after the completion of a query and an answer between the user 10 and the conversation partner 100 explained in FIG. 3, and FIG. 5 And the description of FIG. 3 may be omitted.

5, the interactive automatic question and answer system 1d may include a conversation manager 100, a query response manager 200, a module server 500, and a plug-in server 600. [ The interactive automatic question and answer system 1d may further include a natural language understanding unit 300. [

When the interactive automatic question and answer system 1d receives the natural language query from the user 10 via the network 20, the natural language understanding unit 300 receives the query through the query response manager 200, And a plurality of unit analysis processes for natural language understanding. When the notification receiving unit 210 receives the unit natural language analysis result of at least one unit analysis process among the plurality of unit analysis processes for the query being performed by the natural language understanding unit 300, the module / plug- 620, 630, 640, 650, and 660 of the plug-in server 600 and the plurality of modules 510, 520, and 530 of the module sub 500, Lt; / RTI > and / or plug-in corresponding to < / RTI >

For example, a first module 510, a second module 520, a first plug-in 610, and a second plug-in module 610 of a plurality of modules 510, 520, and 530 and plugins 610, 620, 630, 640, 650, And the second plug-in 620 receive the unit natural language analysis result, and can independently perform the analysis. At this time, the unit natural language analysis results received by the first module 510, the second module 520, the first plug-in 610, and the second plug-in 620 are transmitted to the natural language understanding unit 300, The result may be the same unit natural language analysis result generated in the process, or the unit natural language analysis result generated in the other unit analysis process. The third module 530 and the third through sixth plug-ins 630, 640, 650, and 660 that have not received the unit natural language analysis result are in an idle state without performing analysis, Can be performed.

The modules and / or plug-ins that have been analyzed and generated answers, among the first module 510, the second module 520, the first plug-in 610, and the second plug-in 620 under analysis, The response is sent to the response determination unit 230 without considering whether to perform the analysis of the module and / or the plug-in.

The answer similarity degree determining unit 236 can determine the similarity degree of answers when the number of answers transmitted to the response determining unit 230 is two or more. When the similarity degree of at least two answers among the answers sent to the response decision unit 230 is equal to or greater than a certain standard, at least two answers having a similarity degree equal to or higher than a certain standard level or one of them may be selected as a response to the query.

Also, if there are at least two answers having a similarity degree that is equal to or higher than a certain reference among the transmitted answers, the answer similarity degree determination unit 236 may stop the analysis in the module and / or the plug- .

In some embodiments, the answer ranking determining unit 238 refers to the answer ranking unit 240 to select at least two answers having similarities higher than a certain standard, , A ranking can be selected, and an answer having a high ranking among the answers can be selected as a response to the inquiry.

The additional query determiner 232 is described in detail with reference to FIG. 3, and is omitted.

Referring again to FIG. 2, the interactive automatic question answering system 1a may provide a response to a query using the answer time management unit 234 and the answer similarity determination unit 236 together.

For example, even if the answer similarity determining unit 236 finds at least two answers having a degree of similarity equal to or higher than a predetermined reference even before the elapsed time managed by the answer time managing unit 234 reaches the predetermined second time, It can be provided in response to a query. Or when the answer similarity determining unit 236 finds at least two answers having a degree of similarity equal to or higher than a certain standard, the elapsed time managed by the answer time managing unit 234 becomes the second appointed time, For the answers, the answer ranking unit 238 determines the ranking by referring to the answer ranking unit 240, and selects a response having a high ranking among the answers as a response to the inquiry.

6 is a block diagram illustrating an interactive automatic question and answer system in accordance with an exemplary embodiment of the present invention. The description of FIG. 6 which is the same as the description of FIG. 1 to FIG. 5 may be omitted.

6, the interactive automatic question-answering system 2 includes a dialog manager 100, a question and answer manager 200, a natural language processor, and the like, which receives a query from the user 10 via the network 20 and transmits a response. A natural language generation unit 400, a module server 500, a plug-in server 600, and a domain DB unit 700. As shown in FIG. In some embodiments, the interactive automatic question and answer system 2 may further include a module / plug-in model 900.

The interactive automatic question and answer system 2 is connected to at least one plug-in among the first to sixth plug-ins 610, 620, 630, 640, 650 and 660 of the plug- / DB (800).

The conversation manager 100 includes a communicator 110, a session manager 120, a short term memory 122, a user memory 124, a conversation flow manager 130, a conversation policy 132, 134).

The communicator 110 may provide an interface to allow the conversation to proceed between the user 10 and the interactive automatic question and answer system 2. For example, the communicator 110 may provide an interface such as a chat window.

The session manager 120 may manage the conversation (query, question and answer) with the user for the current query, i.e., the current session, and the previous conversation with the user, i.e., the previous session. The conversation with the user for the current query may be stored in the short memory 122 and the previous conversation with the user, i.e., the conversation (query, question and answer) with the user to the previous query, Lt; / RTI >

For example, if the user 10 makes a query of "What is the weather?", The query may be stored in the short memory 122. If the user 10 then asks " Seoul, " and then the answer is also stored in the short memory 122 have. The query "what is the weather" stored in the short memory 122 and the answer "Seoul" are sent to the query response manager 200 to request a response to the query.

The conversation manager 100 may determine that the previous conversation with the user 10 stored in the user memory 124 has been completed by the user if the information of "Seoul weather" is stored in the previous conversation with the user 10 stored in the user memory 124. [ Quot ;, " Are you talking about the current weather in Seoul? "

The conversation flow manager 130 may refer to the conversation policy 132 and the conversation flow model 134 to cause the user 10 to generate a question in natural language form.

The query response manager 200 includes a notification reception unit 210, a module / plug-in detection unit 220, a module information storage unit 222, a plug-in information storage unit 224, a response determination unit 230, ).

The module server 500 includes a query response module 580 having a plurality of modules 510, 520 and 530 and information on each of a plurality of modules 510, 520 and 530 of the query response module 580 And a module registration unit 590 having a module registration unit 590. The plurality of modules 510, 520, and 530 of the QA module 580 may include, for example, a Factoid Based Question Answering (FBQA) module, a Template Based Question Answering (TBQA) module, a Generated Based Question Answering , An IRQA (Information Retrieval Based Question Answering) module, and the like.

The plug-in server 600 includes a query response plug-in portion 680 having a plurality of plugins 610, 620, 630, 640, 650 and 660 and a plurality of plugins 610, 620, 630, 640, 650, and 660, respectively. The plurality of plug-ins 610, 620, 630, 640, 650 and 660 of the query response plug-in unit 680 may include a mathematical operation plug-in, a distance calculation plug-in, a unit conversion plug- Weather plugins, and the like.

The module information storage unit 222 and the plugin information storage unit 224 respectively store a plurality of modules 510, 520 and 530 and a plurality of plugins 610, 620, and 630 from the module registration unit 590 and the plugin registration unit 690, 630, 640, 650, and 660, respectively. The module / plug-in detection unit 220 selects an appropriate module and / or plug-in corresponding to the received unit natural language analysis result based on the information stored in the module information storage unit 222 and the plug-in information storage unit 224, Natural language analysis results.

The natural language understanding unit 300 includes a semantic role labeling unit (SRL) 310, a morphological analysis unit 320, a syntax analysis unit 322, an entity name analysis unit 324, an intention classification unit 330, 340). 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 322 and the object name analyzer 324 can perform parsing and object name analysis on queries separated by morpheme units. The intention classifying unit 330 and the domain analyzing unit 340 can perform intention classification and domain analysis of a query to which a semantic role is assigned.

In the analysis steps of the semantic roll labeling unit (SRL) 310, morphological analysis unit 320, syntax analysis unit 322, entity name analysis unit 324, intention classification unit 330 and domain analysis unit 340 The result of analysis is called a unit natural language analysis result and includes a semantic roll labeling unit SRL 310, a morphological analysis unit 320, a syntax analysis unit 322, an object name analysis unit 324, an intention classification unit 330 And the domain analysis unit 340 may transmit the unit natural language analysis results generated independently of the other analysis steps to the notification receiving unit 210. [

The notification receiving unit 210 provides the module / plug-in detection unit 220 with the unit natural language analysis result delivered in each analysis step. The module / plug-in detection unit 220 includes a module information storage unit 222 and a plug- 224), the unit natural language analysis result can be transmitted by selecting an appropriate module and / or plug-in corresponding to each unit natural language analysis result. At this time, one unit of natural language analysis result may be transmitted to one module or plug-in, but may be simultaneously transmitted to two or more modules and / or plug-ins.

The plurality of modules 510, 520 and 530 of the query response module 580 refer to at least one domain DB among the plurality of domain DBs 710, 720, 730 and 740 of the domain DB unit 700 , You can create an answer. The plurality of domain DBs 710, 720, 730, and 740 may be knowledge bases in which knowledge data is accumulated. The plurality of domain DBs 710, 720, 730, and 740 may include, for example, a general common knowledge base, a movie domain knowledge base, a music domain knowledge base, a health domain knowledge base, a personal assist domain knowledge base, .

The plurality of plug-ins 610, 620, 630, 640, 650 and 660 of the query response plug-in unit 680 may include a mathematical operation plug-in, a distance calculation plug-in, a unit conversion plug- Weather plugins, and the like. For example, some plugins 610, 630, 650, such as a math operation plugin, a unit conversion plugin, a time date plugin, etc., among a plurality of plugins 610, 620, 630, 640, 650, 660, , 730, and 740). For example, some of the plug-ins 610, 620, 630, 640, 650, 660, some of the plug-ins 620, 640, 660, such as distance calculation plug- 1 service 810, a second service 820, a third service 830, and the like to perform QA. The first service 810, the second service 820 and the third service 830 are connected to the first external DB 812, the second external DB 822 and the third external DB 832, respectively, Or may include. For example, the external service may be a geographical information service, a currency conversion service, a weather service, and the like. The external DB that is connected to the external service or included in the external service may include, for example, a Geo information DB, Weather DB, and the like.

The natural language generation unit 400 includes a response generation unit 410, a module natural language generation rule unit 420, and a plugin natural language generation rule unit 430. The response determining unit 230 and the answer ranking unit 240 can generate a natural-language response by referring to the response generating unit 410. [ A plurality of modules 510, 520, and 530 and a plurality of plugins 610, 620, 630, 640, 650, and 660 refer to the module natural language generation rule unit 420 and the plugin natural language generation rule unit 430, You can create a natural-language answer.

In some embodiments, the plurality of modules 510, 520, 530 and the plurality of plugins 610, 620, 630, 640, 650, 660 refer to the module / plugin model 900, Can be generated.

Module / plug-in model 900 may include query response logic 910, query response execution conditions 920, and natural language generation rules 930. [ The query response logic 910 includes logic for generating an answer from the unit natural language analysis results for the query in the plurality of modules 510, 520, 530 and the plurality of plugins 610, 620, 630, 640, 650, Can be provided. The query response execution condition 920 is a process of generating an answer from the result of the unit natural language analysis on the query in the plurality of modules 510, 520, 530 and the plurality of plugins 610, 620, 630, 640, 650, 660 Can provide the conditions that should be referenced in. The natural language generation rule 930 includes a plurality of modules 510, 520 and 530 and a plurality of plugins 610, 620, 630, 640, 650 and 660, The user can provide a rule for generating a response in the form of a natural language,

7 is a block diagram illustrating a module server and a domain DB unit of the interactive automatic question answering system according to an exemplary embodiment of the present invention in detail.

7, the module server 500 includes a query response module unit 580 having a plurality of modules 510a, 520a, and 530a, and a plurality of modules 510a, 520a, and 560b of the query response module unit 580, And a module registration unit 590 having information on each of the modules 530a and 530a.

The plurality of modules 510a, 520a, and 530a may be FBQA module 510a, TBQA module 520a, and GBQA module 530a, respectively. The FBQA module 510a can automatically generate a SPARQL-type query through keyword-based pattern matching, and then generate an answer by querying the domain DB unit 700, i.e., the knowledge base, using a short-form query. The TBQA module 520a may generate a response by querying the domain DB unit 700, i.e., the knowledge base, after completing the query from the template 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 using syntax analysis, etc., and automatically generates a SPARQL-type query, and generates an answer by querying the domain DB unit 700, i.e., have.

In some embodiments, the query response 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 700 includes a common domain knowledge base 710a, a movie domain knowledge base 720a, a music domain knowledge base 730a, and a health domain knowledge base 740a. Knowledge base.

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 modules 510a, 520a, and 530a of the query response module 580 need only some of the natural language analysis results generated in the plurality of analysis steps of the natural language understanding unit 300 So that analysis of at least some of the plurality of modules 510a, 520a, and 530a can be started even before the plurality of analysis steps of the natural language understanding unit 300 are completed.

8 is a block diagram illustrating in detail an external service / DB connected to a plug-in server and a plug-in server of an interactive automatic question-answering system according to an exemplary embodiment of the present invention.

8, the plug-in server 600 includes a query response plug-in unit 680 having a plurality of plug-ins 610a, 620a, 630a, 640a, 650a, and 660a, and a plurality of query response plug- And a plug-in register 690 having information on each of the plug-ins 610a, 620a, 630a, 640a, 650a, and 660a. Some of the plugins 620a, 640a, and 660a of the plurality of plugins 610a, 620a, 630a, 640a, 650a, and 660a may be connected to the external service / DB 800 through the network 20a.

The plurality of plug-ins 610a, 620a, 630a, 640a, 650a and 660a are connected to a mathematical operation plug-in 610a, a distance calculation plug-in 620a, a unit conversion plug-in 630a, a currency conversion plug-in 640a, 650a, and a weather plugin 660a.

The external service / DB 800 includes, for example, a geographical information service 810a and a Geo information DB 812a, a currency conversion service 820a connected to the Geographic Information service 810a, and a currency exchange information DB 822a , And weather service 830a, and a global weather database 832a associated with or included in the weather service 830a.

The mathematical operation plug-in 610a can itself generate an answer to a query regarding a mathematical operation, such as associating an oblique rule. The distance calculation plug-in 620a may have an API to be associated with the geographic information service 810a and may generate an answer to the query such as distance or path between the two regions through the geographic information service 810a. The unit conversion plug-in 630a can itself generate an answer to the inquiry about the unit conversion. The currency conversion plugin 640a may have an API to be associated with the currency conversion service 820a and may generate an answer to the query such as current exchange rate, exchange rate conversion between different currencies, etc. through the exchange rate conversion service 810a have. The time date plug-in 650a can itself generate responses to queries such as time, date queries, world clocks, volume transitions, and the like. The weather plugin 660a may have an API to be associated with the weather service 830a and may generate responses to queries such as current weather, weather forecasts, past weather, etc. through the weather service 830a.

As such, queries can be generated quickly through plugins 610a, 620a, 630a, 640a, 650a, and 660a for queries that can be generated by themselves or external services without reference to the knowledge base.

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.

1, 1a, 1b, 1c, 1d, 2: Interactive automatic question and answer system, 10: User, 20, 20a: Network, 100: Dialogue manager, 200: Query response manager, 300: Module 500, plug-in server 700, domain DB portion 900, module / plug-in model

Claims (10)

A dialog manager for receiving a query of a user in a natural language form over a network, sending a response, and interactively exchanging questions and answers with the user;
A query response manager managing the user's query and the response thereto;
A natural language understanding unit having a plurality of unit analysis processes for analyzing a query of the user in a natural language form and generating a unit natural language analysis result; And
And a plurality of modules and a plurality of plug-ins that receive the unit natural language analysis results respectively generated in the plurality of unit analysis processes of the natural language understanding unit and independently perform analysis,
The query response manager,
The unit natural language analysis results generated in the plurality of unit analysis processes of the natural language understanding unit are separately received and transmitted to modules or plug-ins corresponding to the unit natural language analysis results of the plurality of modules and the plurality of plug- Lt; / RTI &
An interactive automatic question-answering system for not receiving answers from said plurality of modules and a plurality of plug-ins, or for asking said user through said dialogue manager if additional information is requested. The method according to claim 1,
Wherein the dialogue manager combines the user's answer to the question with the user's query and delivers the combined query to the query response manager. 3. The method of claim 2,
The conversation manager,
A user memory for storing a query of the user for the current query, a short memory storing the question and the user's answer, and a previous query with the user, a question and an answer thereto,
Wherein the query is generated by referring to a previous query with the user stored in the user memory, a question and an answer thereto. The method according to claim 1,
Wherein the query response manager generates the response on the basis of an answer transmitted to a predetermined number of times or a predetermined number of times in a module or plug-in under analysis among the plurality of modules and a plurality of plug- Type automatic question and answer system. 5. The method of claim 4,
Wherein the query response manager stops analysis of a module or a plug-in that has not delivered a response, out of modules or plug-ins being analyzed, when the predetermined time has elapsed. 5. The method of claim 4,
Wherein the query response manager stops analysis of a module or a plug-in that has not delivered a response, out of the module or plug-in being analyzed, when the response is delivered to the predetermined number. 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 the plurality of modules generate an answer with reference to at least one of the plurality of domain DBs,
Wherein the plurality of plug-ins generate an answer without referring to the plurality of domain DBs. 8. The method of claim 7,
Wherein some of the plurality of plug-ins generate their own responses and others are associated with external services to generate responses. The method according to claim 1,
The natural language understanding unit,
A semantic role analyzing unit, a syntax analyzing unit, an entity name analyzing unit, an intention classifying unit, and a domain analyzing unit, each of which performs the plurality of unit analyzing processes,
The unit natural language analysis result generated in the unit analysis process performed in each of the semantic roll labeling unit, the morpheme analysis unit, the syntax analysis unit, the object name analysis unit, the intention classification unit, and the domain analysis unit, To the query response manager in order. The method according to claim 1,
The query response manager,
Wherein the degree of similarity of the responses transmitted within a predetermined time is determined in a module or plug-in under analysis among the plurality of modules and a plurality of plug-ins, and when the similarity degree of the at least two answers is equal to or greater than a certain standard, Based on the answer, generate the response,
And stopping analysis of a module or a plug-in that has not delivered a response, out of modules or plug-ins under analysis, when at least two answers having similarities exceeding the predetermined standard are delivered.

Images