KR101662450B1 - Multi-source hybrid question answering method and system thereof - Google Patents

Multi-source hybrid question answering method and system thereof Download PDF

Info

Publication number
KR101662450B1
KR101662450B1 KR1020150076509A KR20150076509A KR101662450B1 KR 101662450 B1 KR101662450 B1 KR 101662450B1 KR 1020150076509 A KR1020150076509 A KR 1020150076509A KR 20150076509 A KR20150076509 A KR 20150076509A KR 101662450 B1 KR101662450 B1 KR 101662450B1
Authority
KR
South Korea
Prior art keywords
correct answer
knowledge base
query
question
query response
Prior art date
Application number
KR1020150076509A
Other languages
Korean (ko)
Inventor
이근배
박선영
권순철
남대환
한상도
이규송
김병수
Original Assignee
포항공과대학교 산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 포항공과대학교 산학협력단 filed Critical 포항공과대학교 산학협력단
Priority to KR1020150076509A priority Critical patent/KR101662450B1/en
Application granted granted Critical
Publication of KR101662450B1 publication Critical patent/KR101662450B1/en

Links

Images

Classifications

    • G06F17/30976—
    • G06F17/271—
    • G06F17/30734—

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

Disclosed is a multi-source hybrid question answering method and a system thereof capable of outputting suitable answers to questions from users, composed of either a complete sentence or keywords, by utilizing various resources and searching technologies. The multi-source hybrid question answering method is performed in a computing device. The method comprises: a step of categorizing keywords or questions after they are inputted, composed of natural languages; and a step of processing the keywords, outputting a natural language report, processing the questions, and outputting suitable answers. By using an information searching based question answering system and a knowledge based question answering system at the same time and utilizing various technologies to integrated the results obtained from the two systems, shortcomings of using the two systems separately can be compensated.

Description

[0001] MULTI-SOURCE HYBRID QUESTION ANSWERING METHOD AND SYSTEM THEREOF [0002]

The present invention relates to a question and answer system, and more particularly, to a multi-source hybrid query response system in which a question including a question or a keyword consisting of a complete sentence is input from a user, and an appropriate answer is output using various resources and search techniques ≪ / RTI >

The existing question and answer system study aimed to output the correct answer to the question with the complete sentence. The query-response system is a high-level technology that delivers the information that users want in the big data that grow exponentially as the ultimate goal of information retrieval.

However, when it is assumed that a question-answering technology is installed in a home appliance such as a TV, a mobile device, a wearable device, etc., there is an increasing demand for users to get desired information through inquiry while using these devices in daily life . At this time, the user is likely to acquire relevant information not only in a syntactically complete sentence but also in a list of keywords such as " Kim Yu-Na gold medal. &Quot;

In this environment, for example, sophisticated technologies that combine data processing with natural language processing technologies such as Apple's Siri, Google's Now, and Microsoft's personal assistant service, Cortana, have.

In general, the Question Answering System is divided into two categories: an Information Retrieval-based Question Answering System and a Knowledgebase-based Question Answering System. The information retrieval based query response system extracts keywords from the user's query, constructs a query, extracts documents containing correct answers through document retrieval, extracts paragraphs from the documents, and extracts sentences from the paragraphs to find the correct answer . It is difficult to utilize the structured ontology information and it is difficult to solve the ambiguity and synonyms among the correct candidates.

In addition, the knowledge base based query response system is a query response system that finds the correct answer in the structured knowledge base. Recently, with the increase of huge knowledge base such as Yago and Freebase, the importance of open domain query response system using knowledge base is increasing. However, the conventional open domain query response system has a limit in that it can not utilize the context information of the correct candidate in finding the correct answer because there is no document search process.

In order to solve the above problems, an object of the present invention is to provide a multi-source hybrid query response method in which a query in which a question or keyword is composed of a complete sentence is input from a user, and an appropriate answer is output using various resources and a search technique And a system.

Another object of the present invention is to provide a multi-source hybrid query response method and system using a knowledge base based query response system and an information search based query response system simultaneously using an open domain semantic correct answer type detection strategy using a knowledge base ontology .

Another object of the present invention is to provide an efficient and reliable multi-source hybrid query response method and system using a correct answer candidate ranking strategy extracted through knowledge base based query response and information search based query response.

It is still another object of the present invention to provide a method and apparatus for analyzing a query response using resources that integrate ontology information of a text information and a knowledge base, sex resolution information of an entity using a knowledge base, triple extraction result information of text, And to provide a multi-source hybrid query response method and system capable of improving the performance and reliability of the hybrid query response.

According to an aspect of the present invention, there is provided a multi-source hybrid query response method performed in a computing device, the method comprising: discriminating a sentence or a keyword when a sentence or keyword consisting of natural words is input; Outputting a natural language report through the inputting of a sentence, and outputting a correct answer through a query processing on the input of the sentence.

According to another aspect of the present invention, there is provided a computing device for performing a multi-source hybrid query response method, comprising: an input classifier for classifying sentences or keywords when a sentence or a keyword consisting of natural language is input; There is provided a multi-source hybrid query response system including a keyword processing unit for outputting a natural language report through keyword processing and a question processing unit for outputting a correct answer through question processing for the input of a sentence.

Here, the question processor includes a question analyzer, an open domain semantic correct answer type detector, a knowledge base based query response system module, an information search based question and answer system module, and a correct answer candidate ranking module, and a correct answer candidate ranking module Extracts a triple having focus as an attribute, and generates a query using the result of the query analysis used to extract the triple.

In addition, the correct answer candidate ranking module includes a hypothesis and text generator for generating a sentence in which a question is changed into a statement, a focus of the statement is replaced with a candidate for the answer, a sentence or a paragraph including a correct candidate is generated as text, A feature extractor for extracting information constituting the multi-information label database as qualities, and a machine learning-based text box recognition module for hypothesizing and hypothesizing that the hypothesis is true based on the text.

Here, the correct answer candidate ranking module further includes a final score calculation module, wherein the final score calculation module includes a score based on the open domain semantic correct type, a score based on recognition of the text box, The correct answer can be output based on the score based on the semantic similarity of the result.

Here, the multi-source hybrid query response system may further include a database generation unit for building a multiple information label database. The database generation unit includes a natural language processor for processing natural language text, a triple extractor for extracting a triple from a sentence of a natural language text, an object name recognition unit for recognizing the object name in the extracted triple, A matching module, and an entity type extractor for labeling an identifier for the entity in the multiple information label database based on the unique identifier preset in the entity of the knowledge base.

In the case of using the microphone array and the sound source tracking method and system using the coordinate conversion technique according to the embodiment of the present invention as described above, it is necessary to input a question listing a question or keyword composed of a complete sentence from a user, It is possible to efficiently and reliably output an appropriate answer thereto.

In addition, a multi-source hybrid query response method and system using a knowledge base based query response system and an information search based query response system simultaneously using an open domain semantic correct answer type detection strategy using a knowledge base ontology can be provided.

In addition, the performance of the multi - source hybrid query response method and system can be improved by using the correct answer candidate ranking strategy extracted through knowledge base based query response and information search based query response.

In addition, we improve the performance and reliability of query response by using resources that integrate text information and ontology information of knowledge base, sex resolution information of entity using knowledge base, information of triple extraction result of text, and language analysis information There are advantages to be able to.

Furthermore, according to the present embodiment, it is possible to provide a question and answer system capable of processing both user input of a question form and user input of a keyword form. In other words, it is possible to output a report to a keyword inputted by a user by using a natural language generation technique such as robot journalism which is a recent issue, or to output a short answer type answer suitable for a question, Can be provided.

In addition, according to the present embodiment, it is possible to provide a query response system and method using various resources as well as providing an interface that a user can input in various forms. In other words, existing knowledge base based query response systems have limitations in using context information when searching for correct answers, and it is difficult to utilize structured ontology information in information search based query response systems, However, the query response method and system according to the present embodiment can respond appropriately according to whether the input of the user is a keyword or a sentence, as well as a knowledge base based query response system and an information search based query response system Using both resources and strategies to integrate the results obtained from both systems at the same time, it is possible to supplement the limitations of using the knowledge base based query response system and the information search based query response system.

1 is a block diagram of a multi-source hybrid query response system (hereinafter briefly referred to as a hybrid query response system) according to an embodiment of the present invention.
2 is an exemplary view of an input classifier that can be employed in the hybrid query response system of FIG.
3 is a block diagram of an open domain semantic correct answer type detector that can be employed in the hybrid query response system of FIG.
4 is an exemplary view of a database generating unit for constructing the multiple information label database of FIG.
FIG. 5 is a diagram illustrating an example of a process of obtaining context information corresponding to a correct candidate of a knowledge base based query response system module using the processing of the knowledge base based query response system module of FIG. 1. FIG.
FIG. 6 is an exemplary view of another embodiment of the correct candidate ranking module of FIG. 1;
7 is an exemplary view of another embodiment of the correct answer candidate ranking module of FIG.
8 is a block diagram of a multi-source hybrid query response system according to another embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Also, in the present specification, when subscripts of certain characters have different subscripts, other subscripts of subscripts can be displayed in the same form as subscripts for convenience of display.

Unless otherwise defined herein, 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 interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as ideal or overly formal in the sense of the art unless explicitly defined herein Do not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a multi-source hybrid query response system (hereinafter briefly referred to as a hybrid query response system) according to an embodiment of the present invention. 2 is an exemplary view of an input classifier that can be employed in the hybrid query response system of FIG.

The hybrid query response system according to the present embodiment includes an input classification unit 2, a question processing unit 10, and a keyword processing unit 20. The hybrid query response system classifies whether the user input is a keyword or a sentence. When the keyword is classified as a keyword, the hybrid query response system outputs a report about the keyword of the user using the structured knowledge base and the natural language generated template data. Based query response and knowledge base based query response technology at the same time.

In order to use the knowledge base based query response system and the information search based query response system concurrently, the hybrid query response system can use the open domain semantic correct answer type detector using the structured knowledge base ontology, It is possible to use the correct answer candidates ranking module extracted through information retrieval based query response. Also, it can use the text information and ontology information of knowledge base, Named entity disambiguation information using knowledge base, information of triple extraction result, Information, and so on.

The hybrid query response system of the present embodiment will be described in more detail with reference to FIGS. 1 and 2. FIG. Figure 1 illustrates the overall configuration and operational flow of a hybrid query response system. FIG. 2 illustrates a module for automatically classifying questions and keywords.

1 and 2, the input classifier 2 may include a module that learns a binary-classifier model and automatically classifies the input when a user's input is received. For example, if the training data consists of keywords and query statements, the training qualities may include part-of-speech information, uni-grams, bi-grams, Information can be utilized.

When the input of the user is "Tom Cruise Film" (In2), the input classification unit 2 classifies the corresponding user input into the keyword Out2 and transmits the classification result to the keyword processing unit 20. [

The keyword processing unit 20 can process the keyword received from the input classification unit 2 and output a natural language report for the keyword. That is, the keyword processing unit 20 may include a detailed module of the keyword matching module 21, the query generator 22, the triple extractor 23, and the response generator 24 to output an appropriate natural language report for the keyword .

In addition, the keyword processing unit 20 can use the structured knowledge base 18 to output an appropriate natural language report for the keyword. In this embodiment, the structured knowledge base 18 comprises an entity entity < RTI ID = 0.0 >relation; entity> (<object: property: object>). For example, the triple structure, type, or attribute of a keyword that contains the content that Kim's birthday is September 5, 1990 is <Kim yuna; birthDate; 1990-09-05].

In other words, the keyword processing unit 20 may first match the user's keyword with the entity and the attribute of the structured knowledge base, and extract the triple by constructing a query to extract the triple including the detected entity and the attribute. For example, if the input set received from the input classification unit 2 is "Tom Cruise, film", the keyword matching module 21 matches "Tom Cruise, film" with "Tom_Cruise starring" , The query generator 22 generates a query for extracting a triple having "starring" as a property among the triples associated with "Tom_Cruise". And the triple extractor 23 extracts the triple from the structured knowledge base 18 using the query generated in the query generator 22. [ And outputs the natural language report using the template database 25 for generating the natural language of the generated triple. The natural language report can be composed of, for example, "Tom Cruise starred at Interview with the vampire, and Top Gun ...". The template database 25 may include natural language generating resources.

In the above-described keyword processing unit 20, the keyword matching module 21 includes a keyword entity matching module and a keyword property matching module. The keyword entity matching module is constructed by learning by a keyword entity matching trainer based on a keyword entity matching database The keyword property matching module extracts an entity matched with the input keyword in the keyword entity matching model and delivers the extracted entity to the query generator. The keyword property matching module extracts the property matching the input keyword in the keyword property matching model learned by the keyword property matching trainer based on the keyword property matching database And transmit it to the query generator 22. The response generator 24 can extract the natural language corresponding to the triple extracted by the triple extractor 23 connected to the input end of the natural language generating resource 25 and output the natural language report. The natural language generation resource 25 can be prepared by the natural language generation resource generator based on the natural language generation resource database.

On the other hand, if the input of the user is a question such as "Where was Kim yuna born?" (In1), the input classification unit 2 classifies the corresponding user input as a question (Out1) or a question text, 10).

The question processor 10 processes the question received from the input classifier 2 and outputs the correct answer to the question. That is, the query processing unit 20 processes the query sentence to output an appropriate response to the sentence query, and outputs an appropriate correct answer. The detailed module of the question processing unit 10 for processing a user's sentence query includes a question analyzer 11, an open domain semantic correct answer type detector 12, an information search based question and answer system module 13, Response system module 14 and correct answer candidate ranking module 15. The query processing unit 10 may use the multiple information label database 16 in the information search based query response system module 13 and the correct answer candidate ranking module 15. [ The multiple information label database 16 can be constructed by the database creation unit 17 in the off-line.

The question analyzer 11 extracts a focus when a user's query comes in, extracts an object name, extracts a synonym, and analyzes it at a vocabulary level , Analysis at syntactic structure level, analysis at semantic level, and so on. Here, focus refers to the part of the question in the form of "Where was Kim yuna born?", Which is the correct sentence in the form of a statement that the user wants to know if it is replaced with the correct answer or the correct answer candidate.

The question analyzer 11 can use a rule-based method using lexical analysis and parsing results to obtain the focus. For example, first, the question is transformed into a form of "Kim yuna was born in where" in the form of a statement, and detects that "where" is the focus based on the rule. And if you replace "Where" with the correct answer "Bucheon" in this question, you get the same result as "Kim yuna was born in Bucheon".

The results of the query analysis can correspond to the results of the query analyzer 11 analyzing the query, semantic analysis, and parsing at the vocabulary level.

As described above, the focus is a portion to be substituted with the correct answer or the correct candidate, and the type of focus may be the correct type. When the question analysis is finished, the question processor 10 detects the open domain semantic correct answer type by using the analysis result of the question.

The information retrieval based query response system module 13 can search and provide a paragraph judged to contain an appropriate response to an information request or a query using a large-scale text as an information source. That is, the information search based question and answer system module 13 extracts a keyword from a user input query to form a query, extracts a document including correct answers through document search, extracts a paragraph from the document, You can extract and find the correct answer in it.

For reference, the information retrieval based query response system described above and the knowledge base based query response system described below can be referred to as a general knowledge base based query response system and an information retrieval based query response system. The multi-source hybrid query response system according to the present embodiment matches the query analysis result with the attribute of the knowledge base, and matches the recognized entity name with the knowledge base object in the query. We construct a query using the result of syntactic structure analysis of the question, and extract information from the knowledge base using the query. This set of information is the right candidate.

The knowledge base based query response system module 14 is a system for searching for a correct answer to a question in a structured knowledge base 18. Here, knowledge base 18 may include, but is not limited to, a lexical-semantic pattern template module and an analytical parser module. That is, the knowledge base based query response system module 14 according to the present embodiment can be implemented using a general parser without using a semantic parser. In the following, we focus on the case of including an analytic parser as a semantic parser.

The vocabulary-semantic pattern template module includes a vocabulary-semantic pattern dictionary, and may include a pattern matcher module, a template application module, and a format query conversion module. The pattern matcher module checks qualities from a query sentence, The template application module can apply the template corresponding to the matching lexical-semantic pattern to the query sentence, convert it into the actual format query language, and receive and output the response from the knowledge base.

The parser module may include a natural language-attribute dictionary, a combination rule dictionary, an entity name dictionary, etc., and may include a parser, a candidate generation module, a candidate evaluation module, and a format query conversion module. And extracts a response from the knowledge base 18 through a process of extracting formal semantic expressions independently. Here, the parser divides the query sentence into a plurality of non-overlapping parts, and the candidate generating module translates the attribute word into the attribute name and the object name using the natural language-attribute dictionary and the object name dictionary from the divided partial sentences, And the candidate evaluation module can use the candidate evaluation model prepared by the candidate evaluation model trainer based on the query-correct answer database. Then, the formal query conversion module can convert the formal semantic expression evaluated by the candidate evaluation module as the most excellent, into the actual format query language that can be queried in the knowledge base database, and receive and output a response from the database.

The above described parser module can derive purely formal semantic representations from natural language sentences. Using this analytical parser, it is not sensitive to the word order or vocabulary change of the input sentence, so it is possible to grasp the user 's information requirement even in the variation of the sentence, and it is possible to use the vocabulary - It is possible to easily solve the simple information requirement implied by a simple sentence.

On the other hand, the open domain semantic correct answer type detector 12 and the correct candidate ranking module 15 will be described in more detail below with reference to the drawings.

3 is a block diagram of an open domain semantic correct answer type detector that can be employed in the hybrid query response system of FIG.

3, an open domain semantic correct answer type detector (hereinafter, simply referred to as a correct answer type detector) 12 includes a knowledge base use detection unit 12a and a machine learning based detection unit 12b , And detects the semantic correct answer type in the open domain manner. Here the correct answer type is the type of correct answer the user wants to get through the question.

In the conventional information retrieval based query response system, since the correct answer to the query is searched by searching for an answer in a paragraph, a paragraph, a sentence and a sentence in a document, it is possible to filter out the type of the correct answer and other types of objects in the sentence, It is difficult to utilize the structured ontology information and it is difficult to solve the ambiguity between the candidates and to deal with synonyms. Thus, the present embodiment overcomes the limitations of the information search based query response system using the knowledge base when detecting correct answer types. That is, the correct answer type detector 12 according to the present embodiment uses the knowledge base ontology as a correct answer type without using a small set of correct answer types, Can be compared with the type of. In addition, since the semantic correct answer type of the question is extracted using the question analysis result, it can be called an open domain semantic correct answer type detector.

The correct answer type detector 12 of the present embodiment includes an attribute matching module 121, an attribute argument type extraction module 122 and a focus type detector 123 and detects an open domain semantic correct answer type The query analysis result and / or the knowledge base ontology can be used as the correct answer type ontology. An argument type or its attributes can be specified in one or more arguments that are nested in a Command attribute and passed to the script at runtime.

Here, the attribute matching module 121 may measure the semantic similarity between the query analysis result and the knowledge base attribute around the main body of the sentence, and match the semantic predicate of the question with the attribute of the knowledge base. The attribute matching module 121 may match the attribute if the attribute matching exceeds a predetermined threshold based on the knowledge base attribute information of the database 124 and may decide not to match the attribute if the threshold is not exceeded. The attribute argument type extraction module 122 determines the type of entity corresponding to the argument of the attribute based on the attribute of the matched knowledge base in the attribute matching module 121.

Since the focus is a part that is replaced with the correct answer in the query, the focus type detector 123 uses information that the type of focus is the correct answer. That is, the focus type detector 123 can extract the focus and the semantic predicate relation of the question using the result of the question analysis, the relationship between the predetermined focus and the in-question predicate.

For example, in the question "Where was Kim yuna born?", The focus corresponds to "Where" and the relationship between where and the semantic predicate "was born" is a subject-verb relationship. Here, "was born" and the specific attribute "birthPlace" of the knowledge base are mapped to each other, and the argument of the birthPlace attribute corresponds to the place where the subject is located, so the focus is placed in the subject portion of the semantic predicate , So "place" is the type of focus.

On the other hand, the open domain semantic correct answer type detector 12 may fail to match the attribute in the knowledge base use detection unit 12a or the argument type of the attribute may be excessively wide such as "goods "," If the function as a type is not possible, the machine learning based detection unit 12b can be used to detect an open domain semantic correct answer type.

The machine learning based detection unit 12b includes a correct answer type detector 125 and a knowledge base ontology matching module 127. The machine learning based detection unit 12b uses a machine learning based algorithm and uses the result of the question analysis as a quality to generate a correct answer type, The correct answer type is detected. The correct answer type detector 125 may be coupled to the correct answer type detection model learner 1250.

The knowledge base ontology may not be labled in the learning data for the correct answer type detector 125. [ Therefore, the machine learning based detection unit 12b of the present embodiment can previously construct an existing correct answer type ontology mapping table 1270 that maps the correct answer type ontology and the knowledge base ontology that have been used in the past. The existing correct type ontology and knowledge base ontology mapping in the knowledge base ontology matching module 127 can be automatically generated through a comparison between semantic similarity in which vocabulary is vectorized and similar words in two vocabularies, It is possible to minimize the labor force or system load in a way to review.

The next step in query analysis and open domain semantic corrective type detection is query processing in the information retrieval based query response system module and knowledge base based query response system module. Both QA system modules can operate simultaneously. The two question and answer system modules mentioned here can correspond to a general knowledge base based query response system and an information search based query response system.

Here, the knowledge base-based QA system matches the query analysis result with the attribute of the knowledge base, and the recognized entity name in the query can be matched with the knowledge base entity. In addition, the knowledge base based query response system can construct a query using the result of analysis of the structure of the query, and extract information from the knowledge base using the query. This set of information is the right candidate.

The IR-based QA system constructs a query using query analysis results and retrieves the document with this query. Then, it compares the paragraphs of the document with the results of the question analysis, extracts paragraphs that are likely to contain correct answers, and extracts the main sentences that are likely to contain correct answers in the paragraphs. Then extract the objects from the extracted sentences. The extracted objects include sentence, triple, etc., and are candidates for the right answer.

The information search based query response system of this embodiment differs from the existing information search based question and answer system in that it does not use general text but searches a multiple information label database (see 16 in FIG. 1) . Also, the information search based query response system of the present embodiment performs document search, paragraph search, and sentence search in the described order, and at the same time skips the document and paragraph search process using a database indexed on a sentence-by-sentence basis, And triple-unit search can be performed.

4 is an exemplary view of a database generating unit for constructing the multiple information label database of FIG.

4, the database generation unit 17 includes a multiple information extraction unit 170, a multiple information label module 176, and a multiple information generation module 177, The object type extractor 174 and the time information extractor 175 and the multi-information index module 177 includes a natural language processor 171, a triple extractor 172, an entity name recognition and matching module 173, ) May include submodules such as triple unitary index 178, sentence unitary index 179 and document unitary index 180.

The method for constructing the multiple information label database in the database generating unit 17 is a method in which the multiple information extracting unit 170 extracts a plurality of information labels from the general text data using the Natural Language Processing Tool Analysis at the vocabulary level, analysis at the syntactic structure level, analysis at the semantic level, co-reference resolution, and the like can be performed and the results can be labeled through the multiple information label module 176. Here, the multi-information extracting unit 170 may extract a triple from the sentence using the triple extractor 172. [ The triple extractor 172 can separate the long passages. For example, if the triple is extracted from the sentence "Kim was born in 1990 in Bucheon, gyeonggi, and moved to Gunpo when she was six years old. was born in; 1990> <Kim; was born in; Bucheon> <Kim; was born in; Gyeonggi> and <Kim; moved to> Gunpo.

When the multiple information extracting unit 170 is used, an efficient search can be performed by dividing the search target into small units through the triple extractor 172. In addition, the entity name recognition and matching module 173 can identify an entity in a sentence and match the entity in the knowledge base. An entity in the knowledge base has a unique identifier, so the identifier can be labeled.

Also, by using the multiple information extraction unit 170, the knowledge base ontology type of the matched entity can also be labeled through the entity type extractor 174. [ Of course, the type may not be defined. In addition, when the date and time information is not viewed as an object in the knowledge base, the date and time can be separately labeled through the time information extractor 175. [

Using the above-described multi-information extracting unit 170, various information can be labeled for each sentence of the natural language text, and an offset can be also labeled. In this embodiment, the data labeled through the multi-information index module 177 is indexed 180, indexed 179, indexed 178, , And combinations thereof to construct a multiple information label database 16.

According to the present embodiment, the candidates of the correct answers are extracted from the knowledge base based question-and-answer system and the information search based question and answer system, and compared with the correct answer type of the correct candidate type and the correct type of the open domain semantic correct answer type detector, Can be performed.

Here, the type of object, which is the result of the knowledge base based query response system, is provided in the knowledge base ontology and may not be obtained through the index constructed in the system of this embodiment. Here, the knowledge base includes an ontology concept.

In the case of the information retrieval based query response system, the knowledge base ontology is labeled on each entity of the original text (question) by using the aforementioned multi-information label database 16, so that it can be compared with the open domain semantic correct answer type.

5 is a diagram illustrating an example of a process of obtaining context information corresponding to a correct candidate of the knowledge base based query response system module using the processing result of the knowledge base based query response system module of FIG.

5, the correct candidate ranking module 15 of the multi-source hybrid query response system according to the present embodiment includes a query generator 151 and a search module 152, and the processing of the knowledge base based query response system module The context information corresponding to the result can be obtained.

In other words, the result of the knowledge base based QA system can not contain context information as a short-answer type entity. However, context information is needed to rank the correct answer candidates by integrating with the results of the information retrieval based question and answer system. Therefore, the correct answer candidate ranking module 15 searches the multiple information label database based on the combination of the result obtained from the structured knowledge base and the sentence analysis result of the question, regards the search result as the context information of the knowledge base result, Can be performed. In other words, the correct answer candidate ranking module 15 matches the object name recognized in the question with the object of the knowledge base, generates a query using the result of the division structure analysis of the question, and retrieves a sentence obtained from the knowledge base through a query To obtain context information about important keywords or words of a question.

For example, when there is a question "Where was Kim yuna born?" And there is information that the word "Kim yuna" is important in the results of the question analysis, "Kim yuna" You can combine "Bucheon" to create a query and use this query to search for the sentence "Kim yuna Bucheon" to get context information about the result "Bucheon" of the knowledge base based query response system.

To this end, the query generator 151 generates the query using the query analysis result when there is no context information in the result of the knowledge base based query response system. The search module 152 retrieves and / or extracts objects from the multiple information label database 16 using the query generated by the query generator 151. An entity can include sentences, triples extracted from sentences, or various language analysis results for other sentences.

FIG. 6 is an exemplary view of another embodiment of the correct candidate ranking module of FIG. 1;

6, the correct answer candidate ranking module 15 includes a hypothesis and text generator 153, a feature extractor 154, and a machine learning based text box recognition module 155, (Textual Entitlement Recognition), it is possible to generate the correct answer candidate trust score by processing the question and the correct answer candidate. Here, the correct answer candidate confidence score corresponds to the recognition score of the text box.

More specifically, the correct answer candidate ranking module 15 of this embodiment can utilize the recognition technique of the text box for the correct answer candidate ranking. That is, the correct answer candidate ranking module 15 determines whether or not a sentence (for example, Kim) having the answer candidate (Bucheon) and a correct answer candidate (e.g., Kim yuna was born in where) yuna was born in Bucheon ~~). Then, the correct answer candidate ranking module 15 first generates the sentence including the correct answer candidate by replacing the correct answer candidate with the focus through the hypothesis and text generator 153 (Kim yuna was born in Bucheon). This sentence becomes H in the recognition theory of the text box. In addition, the sentence extracted from the correct answer candidate becomes T. This T is used as a kind of evidence to prove that H is true. When H and T are determined, the recognition theory of the text box can be applied in a manner that models are created based on machine learning using the qualities in each of H and T extracted by the feature extractor 154. The final output is a confidence score, where H is true, and the qualities of the query may be various information tagged in the multiple information label database as a result of the query analysis.

In the present embodiment, the recognition theory of the text box determines whether H can be inferred based on T for a given text T (= text) and H (= Hypothesis) through the machine learning based text box recognition module 155 . The recognition theory of this implication can be implemented by the machine learning based textbox recognition module 155 to generate a correct answer candidate confidence score for the given text and hypothesis. Here, the recognition module 155 of the machine learning-based text box can be learned by the recognition recognition model learning machine 150.

7 is an exemplary view of another embodiment of the question processor of FIG.

Referring to FIG. 7, the question processor according to the present embodiment includes a type comparison module 12a, a semantic similarity measurement module 19, and a final score calculation module 156. FIG. The final score calculation module 156 may be included in the correct answer candidate ranking module 15, but is not limited thereto.

The type comparison module 12a compares the detection results from the open domain semantic correct answer type detector with the object types of the correct candidates obtained through the knowledge base based query response system and the object types of the correct candidates obtained through the information search based query response system And outputs the open domain semantic correct answer type score of each correct candidate. If the types match, the open domain semantic correct type score is relatively high, and if the types do not match, the open domain semantic correct type score may be relatively low.

The semantic similarity measurement module 19 measures the semantic similarity based on the extracted sentence and the result of the question analysis, and outputs the sentence score including the correct candidate as the measurement result. Here, the correct answer candidate detection result corresponds to the query response processing result extracted from the multiple information label database by the final score calculation module 15, and the query analysis result may correspond to the query analysis result extracted from the multiple information label database.

The final score calculation module 156 calculates a final score for the correct answer based on the score (open domain semantic correct answer type score) measured through the open domain semantic correct answer type, the recognition score of the text box, and the sentence score including the correct candidate And outputs the final correct answer or final answer list as a calculation result. The recognition score of the textbox may correspond to the recognition result of the machine learning based textbox recognition module (see 155 in FIG. 6) connected to the input of the final score calculation module 156 in the correct candidate ranking module.

8 is a block diagram of a multi-source hybrid query response system according to another embodiment of the present invention.

The hybrid query response system according to the present embodiment may be implemented with at least some functional units of a server apparatus connected to a network or a configuration unit performing functions of these functional units. The hybrid query response system may also be implemented in a computing device that is connected to one or more other user terminals via a wired or wireless network.

8, the hybrid query response system according to the present embodiment includes a processor 100, a memory 110, a database 120, and an input / output device 130, and performs the above-described multi-source hybrid query response method can do.

The processor 100 may include one or more cores, cache memory, and an interface. When the processor 100 has a multi-core structure, a multi-core may refer to integrating two or more independent cores into a single package of a single integrated circuit. And the processor 100 has a single core architecture, a single core may refer to a central processing unit (CPU). The central processing unit (CPU) may be implemented as a system on chip (SOC) in which a micro control unit (MCU) and a peripheral device (integrated circuit for external expansion device) are disposed together, but the present invention is not limited thereto. The core includes registers for storing instructions to be processed, an arithmetic logical unit (ALU) for comparisons, judgments, and arithmetic operations, a control unit (CPU) for internally controlling the CPU to interpret and execute instructions, ), An internal bus, and the like.

In addition, the processor 100 may include, but is not limited to, one or more data processors, image processors, or codecs (CODECs). The data processor, image processor, or codec may be configured separately.

In addition, the processor 100 may have a peripheral interface and a memory interface, in which case the peripheral interface connects the processor 100 to the input / output device 130 and various other peripheral devices, ) And the memory 110 and / or the database 120.

The processor 100 described above may execute data input, data processing, and data output by one or more software programs to perform a multi-source hybrid query response method. In addition, the processor 100 may execute a specific software module (instruction set) stored in the memory 100 to perform various specific functions corresponding to the module. That is, the processor 100 may be able to determine the correct answer to queries of other user terminals in the computing device itself or networked by the modules for the multi-source hybrid query response method implemented by the software modules stored in the memory 100 Natural language reports can be provided.

The memory 110 may include high-speed random access memory and / or non-volatile memory, such as one or more magnetic disk storage devices, one or more optical storage devices, and / or a flash memory. The memory 110 may store software, programs, a set of instructions, or a combination thereof.

Components of the software may include an operating system module, a communication module, a graphics module, a user interface module, a moving picture experts group (MPEG) module, a camera module, one or more application modules, and the like. A module is a set of instructions that can be represented as an instruction set or program.

In addition, the components of the software may include an input classification unit 2, a question processing unit 10, and a keyword processing unit 20 for implementing the multi-source hybrid query response method according to the present embodiment. Each of the input classification unit 2, the question processing unit 10, and the keyword processing unit 20 may include one or more modules. The input classification unit 2, the question processing unit 10, and the keyword processing unit 20 may be executed by the processor 100 to perform the corresponding functions.

The operating system includes built-in operating systems such as MS WINDOWS, LINUX, Darwin, RTXC, UNIX, OS X, iOS, Mac OS, VxWorks, Google OS, Android, And may include various components that control the system operation of the hybrid query response system. The above-described operating system may also include, but is not limited to, a function of performing communication between various hardware devices and software components (modules).

The database 120 includes a plurality of information label databases, a structured knowledge base, a natural language creation resource, a knowledge base attribute information database, a knowledge base ontology database (including an existing correct answer type ontology mapping table), a keyword entity matching database, Keyword property matching database, keyword property matching model database, knowledge base database, natural language resource database, natural language resource, vocabulary-semantic pattern dictionary database, natural language-attribute dictionary database, association rule dictionary database, object name dictionary database, Databases, and the like.

The input / output device 130 includes an external device or a peripheral device such as a keyboard, a mouse, a touch pad, a display device, a touch panel, and an auxiliary storage device. The input / output device 130 may be connected to the processor 100 through an input / output interface.

In addition, the input / output device 130 may include a communication device connected to an external device via a network. A communication device may include one or more wireless communication subsystems. The wireless communication subsystem may include a radio frequency receiver and a transceiver and / or an optical (e.g., infrared) receiver or transceiver. The network may be, for example, a Global System for Mobile Communications (GSM), Enhanced Data GSM Environment (EDGE), Code Division Multiple Access (CDMA), W-Code Division Multiple Access (W-CDMA), Long Term Evolution LTE-Advanced, Orthogonal Frequency Division Multiple Access (OFDMA), WiMax, Wireless Fidelity (Wi-Fi), Bluetooth and the like.

On the other hand, in the present embodiment, the components of the hybrid query response system can be, but are not limited to, functional blocks or modules mounted on the computing device. The above-described components are stored in a computer-readable medium (recording medium) in the form of software for implementing a series of functions (hybrid query response method) performed by them, or transmitted to a remote place in a carrier form so as to be implemented in various computing devices . The computer readable medium may be mounted on a plurality of computer devices or a cloud system connected via a network, and at least one of the plurality of computer devices or the cloud system may perform a hybrid query response method to a memory or a predetermined storage device Or in source code form.

That is, the computer-readable medium may be embodied in a form of program commands, data files, data structures, or the like, alone or in combination, but is not limited thereto. The program recorded on the computer-readable medium may be those specially designed and constructed for the method or system of this embodiment, or may include those known and available to those skilled in the computer software.

The computer-readable medium may also include a hardware device specifically configured to store and execute program instructions, such as a ROM, a RAM, a flash memory, and the like. Program instructions may include machine language code such as those produced by a compiler, as well as high-level language code that may be executed by a computer using an interpreter or the like. The hardware device may be configured to operate with at least one software module to perform the hybrid query response method of the present embodiment, and vice versa.

According to the embodiments described above, by using various strategies for using the information retrieval based query response system and the knowledge base based query response system at the same time and integrating the obtained results, the knowledge base based query response system and the information retrieval based query response It is possible to overcome the limitations of using each system.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the present invention as defined by the following claims It can be understood that

Claims (20)

A multi-source hybrid query response method performed on a computing device,
Discriminating the sentence or the keyword when a sentence or keyword consisting of a natural language is input;
Outputting a natural language report through keyword processing of the input of the keyword; And
And outputting a correct answer through a query process on the input of the sentence,
The step of outputting the correct answer may include:
Extracting focus from the question, and
Detecting an open domain semantic correct answer type using a result of a question analysis used to extract the focus,
Here, the focus refers to a portion of a sentence in the form of a sentence to be understood by the user when the answer is replaced with a correct answer,
Wherein the step of detecting the open domain semantic correct answer type comprises the steps of: using a knowledge base ontology as a correct answer type and extracting a semantic correct answer type of the question using the question analysis result; The semantic similarity between the query analysis result and the structured knowledge base attribute based on the main verb of the sentence is measured, the semantic predicate of the question is matched with the attribute of the knowledge base,
The knowledge base ontology is mapped by a previously stored correct answer type ontology and a knowledge base ontology mapping table,
Wherein the mapping between the correct answer type ontology and the knowledge base ontology is such that a correct answer type is automatically generated by comparing semantic similarity obtained by vectorizing vocabularies and similar words between two vocabularies. The method according to claim 1,
Wherein the outputting of the natural language report comprises:
Generating a query based on the keyword;
Extracting a triple from a structured knowledge base using the query; And
Generating a natural language response for the triple;
Multi - source hybrid query response method. delete delete delete delete The method according to claim 1,
If the failure to detect the primary correct answer type in the step of detecting the open domain semantic correct answer type is detected using a machine learning algorithm performed after the step of detecting the open domain semantic correct answer type, Further comprising the steps of: The method according to claim 1,
The step of outputting the correct answer may include: after the step of detecting an open domain semantic correct answer type, matching the query analysis result with a knowledge base attribute through a knowledge base based query response system module, Based query response processing for matching a query with an entity of a base and constructing a query using the result of the syntax analysis of the query and extracting information from the knowledge base using the query, Question and answer method. The method of claim 8,
Wherein the information extracted from the knowledge base using the query includes context information as a result obtained by inputting the query into a multiple information label database, and the context information includes a triple extracted from a sentence or a sentence. How to respond. The method of claim 8,
Wherein the step of outputting the correct answer comprises: constructing a query using the query analysis result after detecting an open domain semantic correct answer type; searching the document from a multiple information label database using the query; Extracting paragraphs having a relatively high possibility of correct answer by comparing the paragraphs with the result of the question analysis, extracting a main sentence having a relatively high possibility that correct answers are included in the extracted paragraphs, and extracting objects from the sentence Further comprising an information search based query response processing step. The method of claim 10,
The information search based query response processing step performs a document search, a paragraph search and a sentence search in the order described in the search of the multiple information label database, skips the document and paragraph search process using a database indexed on a sentence level basis, A method for multi-source hybrid query response that performs a search of units and a search of triples. The method of claim 10,
Wherein the knowledge base based query response processing step and the information search based query response processing step are performed concurrently and at least one of the set of information and the entities is a correct candidate. The method of claim 12,
The step of outputting the correct answer further includes a correct answer candidate ranking step using recognition of a text box after the knowledge base based query response processing step and the information search based query response processing step,
Wherein the recognition of the text box is performed by replacing the question with a statement, setting a sentence in which the focus of the question is replaced with a correct answer as a hypothesis, and writing a sentence in which the correct answer candidate is replaced with a focus, Wherein the qualities include information tagged in a multi-information label database as a result of a query analysis, wherein the qualities are based on a machine learning basis using qualities in a multi-source hybrid query response method. 14. The method of claim 13,
The correct answer candidate ranking step may include a score obtained by recognizing the score of the open domain semantic correct answer type, a score obtained by recognizing the text box, a sentence extracted from the correct answer candidates, and a score obtained by measuring semantic similarity of the result of the question analysis. A multi-source hybrid query response method that outputs correct answers. A computing device for performing a multi-source hybrid query response method,
An input classifier for classifying the question or keyword when a question or keyword consisting of natural language is input;
A keyword processing unit for outputting a natural language report through a query response process for the keyword; And
And a question processor for outputting a correct answer through a query response process for the question,
The question processing unit,
A question analyzer for performing lexical analysis and classification analysis and extracting focus when the question is received;
An open domain semantic correct answer type detector for detecting an open domain semantic correct answer type based on a question analysis result used to extract the focus;
A knowledge base based query response system module for detecting an entity or a correct answer candidate in a structured knowledge base using the result of the question analysis as a qualification;
An information search based query response system module for measuring semantic similarity between a query analysis result and a knowledge base attribute based on the main verb of the question and matching a semantic predicate of the query to an attribute of a knowledge base; And
An open domain semantic correct answer type score obtained by comparing the detection result of the open domain semantic correct answer type detector with a type found in a multiple information label database and a knowledge base and a recognition result of a text box in the information search based query response system module And a correct answer candidate ranking module for outputting the correct answer using a recognition score of the obtained text box and a sentence score including a correct answer candidate obtained based on the result of the question analysis and the result of detection of the correct answer candidate. 16. The method of claim 15,
The keyword processing unit,
A query generator for generating a query based on the keyword;
A triple extractor for extracting a triple from a structured knowledge base using the query; And
And a response generator for generating a natural language response using the triple,
Multi - source hybrid query response system. delete 16. The method of claim 15,
Wherein the knowledge base based query response system module and the information based query response system module are performed simultaneously. 16. The method of claim 15,
Wherein the open domain semantic correct answer type detector uses a knowledge base ontology for detection of a correct answer type, the knowledge base ontology is mapped by a previously stored correct answer type ontology and a knowledge base ontology mapping table, The mapping of the knowledge base ontology automatically generates the correct answer type by comparing semantic similarity vectorized vocabularies and similar words of two vocabularies. The method of claim 19,
Wherein the open domain semantic correct answer type detector comprises a knowledge base use detection unit and a machine learning based detection unit,
Wherein the knowledge base utilization detecting unit comprises:
An attribute matching module for measuring semantic similarity between attributes of a structured knowledge base having a semantic predicate of the question and matching a semantic predicate of the question to an attribute of the knowledge base;
An attribute argument type extraction module for determining a type of an entity corresponding to an argument of the attribute as a correct answer type of the question according to an attribute of the knowledge base;
And a focus type detector for determining a type of a correct answer candidate using at least one of a relationship between a predetermined focus and a predicate and a question analysis result,
Wherein the machine learning-
The attribute matching failure in the knowledge base utilizing detection unit or the argument type of the attribute does not exceed a threshold,
A correct answer type detector for determining a correct answer type using the question analysis result,
And a knowledge base ontology matching module that outputs an open domain semantic correct answer type using a predetermined correct answer type ontology and a knowledge base ontology mapping table based on the type of correct answer detected by the correct answer type detector.
Multi - source hybrid query response system.
KR1020150076509A 2015-05-29 2015-05-29 Multi-source hybrid question answering method and system thereof KR101662450B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020150076509A KR101662450B1 (en) 2015-05-29 2015-05-29 Multi-source hybrid question answering method and system thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020150076509A KR101662450B1 (en) 2015-05-29 2015-05-29 Multi-source hybrid question answering method and system thereof

Publications (1)

Publication Number Publication Date
KR101662450B1 true KR101662450B1 (en) 2016-10-05

Family

ID=57153837

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020150076509A KR101662450B1 (en) 2015-05-29 2015-05-29 Multi-source hybrid question answering method and system thereof

Country Status (1)

Country Link
KR (1) KR101662450B1 (en)

Cited By (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106874441A (en) * 2017-02-07 2017-06-20 腾讯科技（上海）有限公司 Intelligent answer method and apparatus
KR20180058253A (en) * 2016-11-23 2018-06-01 한국전자통신연구원 Data processing apparatus and method for merging deterministic and non-deterministic knowledge information processing
CN108170780A (en) * 2017-12-26 2018-06-15 北京邦邦共赢网络科技有限公司 A kind of the problem of self-service question and answer matching process and device
KR101928060B1 (en) 2017-12-01 2018-12-11 사회복지법인 삼성생명공익재단 Method and System for Expanding Ideas and Computer Readable Recording Medium Thereof
CN109522394A (en) * 2018-10-12 2019-03-26 北京奔影网络科技有限公司 Knowledge base question and answer system and method for building up
CN109858007A (en) * 2017-11-30 2019-06-07 上海智臻智能网络科技股份有限公司 Semantic analysis answering method and device, computer equipment and storage medium
KR20190081112A (en) * 2017-12-29 2019-07-09 주식회사 헤르스 Correct answer query management system for chemical safety management using deep learning technology
CN110019838A (en) * 2017-12-25 2019-07-16 上海智臻智能网络科技股份有限公司 Intelligent Answer System and intelligent terminal
CN110147358A (en) * 2017-11-22 2019-08-20 上海智臻智能网络科技股份有限公司 The building method and construction system of automatic question answering knowledge base
CN110245216A (en) * 2019-06-13 2019-09-17 出门问问信息科技有限公司 For the semantic matching method of question answering system, device, equipment and storage medium
KR20200014046A (en) * 2018-07-31 2020-02-10 주식회사 포티투마루 Device and Method for Machine Reading Comprehension Question and Answer
CN110858100A (en) * 2018-08-22 2020-03-03 北京搜狗科技发展有限公司 Method and device for generating association candidate words
KR20200068105A (en) * 2018-11-28 2020-06-15 주식회사 솔트룩스 System of providing documents for machine reading comprehension and question answering system including the same
KR20200083053A (en) * 2018-12-31 2020-07-08 (주) 스펠릭스 Method for providing post-processing for improving the accuracy of named-entity recognition, and server using the same
KR20200094627A (en) * 2019-01-30 2020-08-07 베이징 바이두 넷컴 사이언스 앤 테크놀로지 코., 엘티디. Method, apparatus, device and medium for determining text relevance
KR20200101525A (en) * 2019-01-31 2020-08-28 주식회사 카카오 Method and system for analysis of natural language query
CN111651560A (en) * 2020-05-29 2020-09-11 北京百度网讯科技有限公司 Method and device for configuring questions, electronic equipment and computer readable medium
WO2020206243A1 (en) * 2019-04-03 2020-10-08 RELX Inc. Systems and methods for dynamically displaying a user interface of an evaluation system processing textual data
CN111949779A (en) * 2020-07-29 2020-11-17 交控科技股份有限公司 Intelligent rail transit response method and system based on knowledge graph
CN112037905A (en) * 2020-07-16 2020-12-04 朱卫国 Medical question answering method, equipment and storage medium
CN112052311A (en) * 2019-05-20 2020-12-08 天津科技大学 Short text question-answering method and device based on word vector technology and knowledge graph retrieval
CN112434158A (en) * 2020-11-13 2021-03-02 北京创业光荣信息科技有限责任公司 Enterprise label acquisition method and device, storage medium and computer equipment
CN112487168A (en) * 2020-12-11 2021-03-12 润联软件系统(深圳)有限公司 Semantic questioning and answering method and device for knowledge graph, computer equipment and storage medium
CN112507105A (en) * 2021-01-26 2021-03-16 王三山 Multi-mode intelligent question-answering system and method based on WeChat public number
CN112507100A (en) * 2020-12-18 2021-03-16 北京百度网讯科技有限公司 Method and device for updating question-answering system
CN112685548A (en) * 2020-12-31 2021-04-20 中科讯飞互联(北京)信息科技有限公司 Question answering method, electronic device and storage device
CN113742469A (en) * 2021-09-03 2021-12-03 科讯嘉联信息技术有限公司 Pipeline processing and ES storage based question-answering system construction method
CN113806500A (en) * 2021-02-09 2021-12-17 京东科技控股股份有限公司 Information processing method and device and computer equipment
CN113836280A (en) * 2021-08-31 2021-12-24 上海欣能信息科技发展有限公司 Power grid operation detection voice question-answer method and system based on knowledge graph technology
CN113946666A (en) * 2021-09-13 2022-01-18 东北大学 Simple question knowledge base question-answering method based on domain perception
KR20220027273A (en) * 2019-09-03 2022-03-07 미쓰비시덴키 가부시키가이샤 Information processing apparatus, computer readable recording medium and information processing method
WO2022107989A1 (en) * 2020-11-23 2022-05-27 숭실대학교산학협력단 Method and device for completing knowledge by using relation learning between query and knowledge graph
CN114996429A (en) * 2022-06-29 2022-09-02 支付宝(杭州)信息技术有限公司 Method, system, apparatus and medium for automatic question answering
CN116303981A (en) * 2023-05-23 2023-06-23 山东森普信息技术有限公司 Agricultural community knowledge question-answering method, device and storage medium
CN117131181A (en) * 2023-10-24 2023-11-28 国家电网有限公司 Construction method of heterogeneous knowledge question-answer model, information extraction method and system
CN117171308A (en) * 2023-07-28 2023-12-05 至本医疗科技（上海）有限公司 Method, device and medium for generating scientific research data analysis response information
CN117290694A (en) * 2023-11-24 2023-12-26 北京并行科技股份有限公司 Question-answering system evaluation method, device, computing equipment and storage medium
CN117407515A (en) * 2023-12-15 2024-01-16 湖南三湘银行股份有限公司 Answer system based on artificial intelligence
CN117648429A (en) * 2024-01-30 2024-03-05 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) Question-answering method and system based on multi-mode self-adaptive search type enhanced large model
CN113806500B (en) * 2021-02-09 2024-05-28 京东科技控股股份有限公司 Information processing method, device and computer equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110020462A (en) * 2009-08-24 2011-03-03 송도규 System and method for intelligent searching and question-answering
KR20120053253A (en) * 2010-11-17 2012-05-25 주식회사 케이티 System and method for hybrid semantic searching service
KR20120087346A (en) * 2011-01-25 2012-08-07 김일표 System and method for providing information between coperations and customers

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110020462A (en) * 2009-08-24 2011-03-03 송도규 System and method for intelligent searching and question-answering
KR20120053253A (en) * 2010-11-17 2012-05-25 주식회사 케이티 System and method for hybrid semantic searching service
KR20120087346A (en) * 2011-01-25 2012-08-07 김일표 System and method for providing information between coperations and customers

Cited By (65)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20180058253A (en) * 2016-11-23 2018-06-01 한국전자통신연구원 Data processing apparatus and method for merging deterministic and non-deterministic knowledge information processing
KR102091240B1 (en) 2016-11-23 2020-03-20 한국전자통신연구원 Data processing apparatus and method for merging deterministic and non-deterministic knowledge information processing
US11144833B2 (en) 2016-11-23 2021-10-12 Electronics And Telecommunications Research Institute Data processing apparatus and method for merging and processing deterministic knowledge and non-deterministic knowledge
CN106874441A (en) * 2017-02-07 2017-06-20 腾讯科技（上海）有限公司 Intelligent answer method and apparatus
CN106874441B (en) * 2017-02-07 2024-03-05 腾讯科技（上海）有限公司 Intelligent question-answering method and device
CN110147358A (en) * 2017-11-22 2019-08-20 上海智臻智能网络科技股份有限公司 The building method and construction system of automatic question answering knowledge base
CN110147358B (en) * 2017-11-22 2024-05-17 上海智臻智能网络科技股份有限公司 Construction method and construction system of automatic question-answering knowledge base
CN109858007B (en) * 2017-11-30 2024-02-02 上海智臻智能网络科技股份有限公司 Semantic analysis question-answering method and device, computer equipment and storage medium
CN109858007A (en) * 2017-11-30 2019-06-07 上海智臻智能网络科技股份有限公司 Semantic analysis answering method and device, computer equipment and storage medium
KR101928060B1 (en) 2017-12-01 2018-12-11 사회복지법인 삼성생명공익재단 Method and System for Expanding Ideas and Computer Readable Recording Medium Thereof
CN110019838A (en) * 2017-12-25 2019-07-16 上海智臻智能网络科技股份有限公司 Intelligent Answer System and intelligent terminal
CN108170780A (en) * 2017-12-26 2018-06-15 北京邦邦共赢网络科技有限公司 A kind of the problem of self-service question and answer matching process and device
KR102005525B1 (en) * 2017-12-29 2019-07-30 (주)헤르스 Correct answer query management system for chemical safety management using deep learning technology
KR20190081112A (en) * 2017-12-29 2019-07-09 주식회사 헤르스 Correct answer query management system for chemical safety management using deep learning technology
KR20200014046A (en) * 2018-07-31 2020-02-10 주식회사 포티투마루 Device and Method for Machine Reading Comprehension Question and Answer
KR102088357B1 (en) * 2018-07-31 2020-03-12 주식회사 포티투마루 Device and Method for Machine Reading Comprehension Question and Answer
CN110858100B (en) * 2018-08-22 2023-10-20 北京搜狗科技发展有限公司 Method and device for generating association candidate words
CN110858100A (en) * 2018-08-22 2020-03-03 北京搜狗科技发展有限公司 Method and device for generating association candidate words
CN109522394A (en) * 2018-10-12 2019-03-26 北京奔影网络科技有限公司 Knowledge base question and answer system and method for building up
KR102130779B1 (en) * 2018-11-28 2020-07-08 주식회사 솔트룩스 System of providing documents for machine reading comprehension and question answering system including the same
KR20200068105A (en) * 2018-11-28 2020-06-15 주식회사 솔트룩스 System of providing documents for machine reading comprehension and question answering system including the same
KR20200083053A (en) * 2018-12-31 2020-07-08 (주) 스펠릭스 Method for providing post-processing for improving the accuracy of named-entity recognition, and server using the same
KR102153127B1 (en) 2018-12-31 2020-09-07 (주) 스펠릭스 Method for providing post-processing for improving the accuracy of named-entity recognition, and server using the same
KR20200094627A (en) * 2019-01-30 2020-08-07 베이징 바이두 넷컴 사이언스 앤 테크놀로지 코., 엘티디. Method, apparatus, device and medium for determining text relevance
KR102564144B1 (en) 2019-01-30 2023-08-08 베이징 바이두 넷컴 사이언스 앤 테크놀로지 코., 엘티디. Method, apparatus, device and medium for determining text relevance
KR20200101525A (en) * 2019-01-31 2020-08-28 주식회사 카카오 Method and system for analysis of natural language query
KR102150908B1 (en) * 2019-01-31 2020-09-03 주식회사 카카오 Method and system for analysis of natural language query
WO2020206243A1 (en) * 2019-04-03 2020-10-08 RELX Inc. Systems and methods for dynamically displaying a user interface of an evaluation system processing textual data
US11520990B2 (en) 2019-04-03 2022-12-06 RELX Inc. Systems and methods for dynamically displaying a user interface of an evaluation system processing textual data
US11972217B2 (en) 2019-04-03 2024-04-30 RELX Inc. Systems and methods for dynamically displaying a user interface of an evaluation system processing textual data
CN112052311A (en) * 2019-05-20 2020-12-08 天津科技大学 Short text question-answering method and device based on word vector technology and knowledge graph retrieval
CN110245216A (en) * 2019-06-13 2019-09-17 出门问问信息科技有限公司 For the semantic matching method of question answering system, device, equipment and storage medium
KR102473788B1 (en) 2019-09-03 2022-12-02 미쓰비시덴키 가부시키가이샤 Information processing device, computer readable recording medium and information processing method
KR20220027273A (en) * 2019-09-03 2022-03-07 미쓰비시덴키 가부시키가이샤 Information processing apparatus, computer readable recording medium and information processing method
CN111651560A (en) * 2020-05-29 2020-09-11 北京百度网讯科技有限公司 Method and device for configuring questions, electronic equipment and computer readable medium
CN111651560B (en) * 2020-05-29 2023-08-29 北京百度网讯科技有限公司 Method and device for configuring problems, electronic equipment and computer readable medium
CN112037905A (en) * 2020-07-16 2020-12-04 朱卫国 Medical question answering method, equipment and storage medium
CN111949779A (en) * 2020-07-29 2020-11-17 交控科技股份有限公司 Intelligent rail transit response method and system based on knowledge graph
CN112434158B (en) * 2020-11-13 2024-05-28 海创汇科技创业发展股份有限公司 Enterprise tag acquisition method, enterprise tag acquisition device, storage medium and computer equipment
CN112434158A (en) * 2020-11-13 2021-03-02 北京创业光荣信息科技有限责任公司 Enterprise label acquisition method and device, storage medium and computer equipment
WO2022107989A1 (en) * 2020-11-23 2022-05-27 숭실대학교산학협력단 Method and device for completing knowledge by using relation learning between query and knowledge graph
CN112487168B (en) * 2020-12-11 2024-03-08 华润数字科技有限公司 Semantic question-answering method and device of knowledge graph, computer equipment and storage medium
CN112487168A (en) * 2020-12-11 2021-03-12 润联软件系统(深圳)有限公司 Semantic questioning and answering method and device for knowledge graph, computer equipment and storage medium
CN112507100A (en) * 2020-12-18 2021-03-16 北京百度网讯科技有限公司 Method and device for updating question-answering system
CN112507100B (en) * 2020-12-18 2023-12-22 北京百度网讯科技有限公司 Update processing method and device of question-answering system
CN112685548A (en) * 2020-12-31 2021-04-20 中科讯飞互联(北京)信息科技有限公司 Question answering method, electronic device and storage device
CN112685548B (en) * 2020-12-31 2023-09-08 科大讯飞(北京)有限公司 Question answering method, electronic device and storage device
CN112507105A (en) * 2021-01-26 2021-03-16 王三山 Multi-mode intelligent question-answering system and method based on WeChat public number
CN113806500B (en) * 2021-02-09 2024-05-28 京东科技控股股份有限公司 Information processing method, device and computer equipment
CN113806500A (en) * 2021-02-09 2021-12-17 京东科技控股股份有限公司 Information processing method and device and computer equipment
CN113836280A (en) * 2021-08-31 2021-12-24 上海欣能信息科技发展有限公司 Power grid operation detection voice question-answer method and system based on knowledge graph technology
CN113742469B (en) * 2021-09-03 2023-12-15 科讯嘉联信息技术有限公司 Method for constructing question-answering system based on Pipeline processing and ES storage
CN113742469A (en) * 2021-09-03 2021-12-03 科讯嘉联信息技术有限公司 Pipeline processing and ES storage based question-answering system construction method
CN113946666A (en) * 2021-09-13 2022-01-18 东北大学 Simple question knowledge base question-answering method based on domain perception
CN114996429A (en) * 2022-06-29 2022-09-02 支付宝(杭州)信息技术有限公司 Method, system, apparatus and medium for automatic question answering
CN116303981A (en) * 2023-05-23 2023-06-23 山东森普信息技术有限公司 Agricultural community knowledge question-answering method, device and storage medium
CN117171308A (en) * 2023-07-28 2023-12-05 至本医疗科技（上海）有限公司 Method, device and medium for generating scientific research data analysis response information
CN117131181B (en) * 2023-10-24 2024-04-05 国家电网有限公司 Construction method of heterogeneous knowledge question-answer model, information extraction method and system
CN117131181A (en) * 2023-10-24 2023-11-28 国家电网有限公司 Construction method of heterogeneous knowledge question-answer model, information extraction method and system
CN117290694B (en) * 2023-11-24 2024-03-15 北京并行科技股份有限公司 Question-answering system evaluation method, device, computing equipment and storage medium
CN117290694A (en) * 2023-11-24 2023-12-26 北京并行科技股份有限公司 Question-answering system evaluation method, device, computing equipment and storage medium
CN117407515A (en) * 2023-12-15 2024-01-16 湖南三湘银行股份有限公司 Answer system based on artificial intelligence
CN117407515B (en) * 2023-12-15 2024-04-30 湖南三湘银行股份有限公司 Answer system based on artificial intelligence
CN117648429A (en) * 2024-01-30 2024-03-05 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) Question-answering method and system based on multi-mode self-adaptive search type enhanced large model
CN117648429B (en) * 2024-01-30 2024-04-30 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) Question-answering method and system based on multi-mode self-adaptive search type enhanced large model

Similar Documents

Publication Publication Date Title
KR101662450B1 (en) Multi-source hybrid question answering method and system thereof
JP5936698B2 (en) Word semantic relation extraction device
Yang et al. Joint relational embeddings for knowledge-based question answering
JP7232831B2 (en) Retrieval of corroborative evidence for complex answers
US11816441B2 (en) Device and method for machine reading comprehension question and answer
US20180341871A1 (en) Utilizing deep learning with an information retrieval mechanism to provide question answering in restricted domains
US9280535B2 (en) Natural language querying with cascaded conditional random fields
US9286290B2 (en) Producing insight information from tables using natural language processing
KR102316063B1 (en) Method and apparatus for identifying key phrase in audio data, device and medium
US10169703B2 (en) System and method for analogy detection and analysis in a natural language question and answering system
CN101878476B (en) Machine translation for query expansion
CN111914568A (en) Method, device and equipment for generating text modifying sentence and readable storage medium
CN111046656B (en) Text processing method, text processing device, electronic equipment and readable storage medium
US10503830B2 (en) Natural language processing with adaptable rules based on user inputs
Xie et al. Topic enhanced deep structured semantic models for knowledge base question answering
US20220405484A1 (en) Methods for Reinforcement Document Transformer for Multimodal Conversations and Devices Thereof
US11893537B2 (en) Linguistic analysis of seed documents and peer groups
CN105760462A (en) Man-machine interaction method and device based on associated data query
US20220414463A1 (en) Automated troubleshooter
CN112507089A (en) Intelligent question-answering engine based on knowledge graph and implementation method thereof
CN111859858A (en) Method and device for extracting relationship from text
Chen et al. A synergistic framework for geographic question answering
US20220147719A1 (en) Dialogue management
CN113157887A (en) Knowledge question-answering intention identification method and device and computer equipment
US20210165791A1 (en) Method and apparatus for building a paraphrasing model for question-answering

Legal Events

Date Code Title Description
GRNT Written decision to grant