KR101988396B1 - System for generting query to knowledge base from natural language question and ranking resources and question answering system including the same - Google Patents

Abstract

According to an exemplary embodiment of the present invention, provided is a question and answer system, which comprises: a question pattern analysis unit configured to generate at least one question pattern arranged in accordance with a ranking by mapping tokens extracted from a natural language question to ontology components based on meta-knowledge data; and a query generation unit configured to generate at least one query on the knowledge base based on a question pattern template from the at least one question pattern. The meta-knowledge data may include meta-information of the ontology components in the knowledge base.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for generating a query and a resource ranking for a knowledge base from a natural language query and a query response system including the same,

The technical idea of the present invention relates to a knowledge base, and more particularly, to a system and a method for generating a query on a knowledge base.

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

A knowledge base may be constructed that stores knowledge data and provides stored knowledge data. In order to provide the knowledge desired by the user among the knowledge (or knowledge data) included in the knowledge base, the knowledge base can support a query. In order to acquire the desired knowledge from the knowledge base, the expert can directly write an appropriate query to the knowledge base, while the non-expert can not easily write the query to the general user.

The query response system can generate a response to the user's query based on the knowledge base. For example, the query response system can generate a response by searching the knowledge base for relevant knowledge according to the natural language query received from the user. In such a QA system, a query to a knowledge base generated from a natural language query can influence the quality of the response provided by the QA system.

The technical idea of the present invention provides a system and method for generating a query for a knowledge base.

In order to achieve the above object, a system according to the technical idea of the present invention maps tokens extracted from a natural language query to ontology components based on meta-knowledge data, And a query generator configured to generate at least one query for the knowledge base from at least one query pattern based on the query pattern template, wherein the query generator is configured to generate at least one query pattern, The meta-knowledge data may include meta information of the ontology components in the knowledge base.

According to an exemplary embodiment of the present invention, the meta-knowledge data may include meta information of an entity including at least one of an importance of an entity, a link count, an attribute list, and a class list, The department may include an entity evaluator configured to calculate a score of the at least one entity based on meta information of the at least one entity corresponding to the token.

According to an exemplary embodiment of the present invention, the entity evaluation unit may calculate the score of the at least one entity so as to be proportional to at least one of the importance and the link count.

According to an exemplary embodiment of the present invention, the entity evaluating unit calculates the score of at least one entity so as to increase when the attribute included in the attribute list or the token corresponding to the class included in the class list is extracted can do.

According to an exemplary embodiment of the present invention, the meta-knowledge data may include meta information of an attribute including at least one of hierarchical information of attributes, domain information, and range information, and the query pattern analysis unit may correspond to a token And an attribute evaluation unit configured to calculate a score of the at least one attribute based on the meta information of at least one attribute.

According to an exemplary embodiment of the present invention, the attribute evaluation unit may calculate the score of the at least one attribute such that the attribute of the higher level has a lower score, based on the hierarchical information of the attributes.

According to an exemplary embodiment of the present invention, the attribute evaluation unit may have a token corresponding to a class of an associated domain or range of at least one attribute of extracted tokens, based on at least one of domain information and range information of attributes , The score of at least one attribute can be calculated.

According to an exemplary embodiment of the present invention, the meta-knowledge data may include meta information of a class including hierarchical information of classes, and the query pattern analyzing unit may include meta information of at least one class corresponding to the token And a class evaluation unit configured to calculate a score of at least one class based on the evaluation result.

According to an exemplary embodiment of the present invention, the class evaluation unit may calculate the score of at least one class so that the higher-level class has a lower score, based on the hierarchical information of the classes.

According to an exemplary embodiment of the present invention, the query pattern analyzing unit includes a query pattern rank unit configured to determine a ranking of at least one query pattern, based on the scores of the ontology components included in the at least one query pattern .

According to the system and method according to the technical idea of the present invention, a query can be generated for a knowledge base that matches the intention of a natural language query.

In addition, according to the system and method according to the technical idea of the present invention, a query and a response can be quickly generated with respect to a user's query due to pre-established meta-knowledge data.

Further, according to the system and method according to the technical idea of the present invention, the meta-knowledge data can be automatically updated according to the update of the knowledge base.

1 is a block diagram illustrating a query response system and its input / output relationship according to an exemplary embodiment of the present invention.
Figure 2 is a block diagram illustrating an example of the query pattern analyzer of Figure 1 in accordance with an exemplary embodiment of the present disclosure.
Figure 3 illustrates an example of at least one token list of Figure 2 in accordance with an exemplary embodiment of the present disclosure.
Figure 4 illustrates an example of entity meta information in Figure 2 in accordance with an exemplary embodiment of the present disclosure.
Figures 5 and 6 illustrate examples of the attribute meta information of Figure 2 in accordance with the exemplary embodiments of the present disclosure.
Figure 7 illustrates an example of at least one query pattern of Figure 2 in accordance with an exemplary embodiment of the present disclosure.
8 is a diagram illustrating an example of pattern template data included in the pattern template database of FIG. 1 according to an exemplary embodiment of the present disclosure.
9 is a flow diagram illustrating a method for generating a query for a knowledge base in accordance with an exemplary embodiment of the present disclosure.
10 is a flow chart illustrating an example of step S40 of FIG. 9 in accordance with an exemplary embodiment of the present disclosure.
11 is a block diagram illustrating a meta-knowledge building system and its input / output relationship in accordance with an exemplary embodiment of the present disclosure;

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 should be construed to have meanings consistent with the meaning in the context of the relevant art and, unless expressly defined in this application, are to be interpreted in an ideal or overly formal sense It does not.

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. Also, a "system" or "database" herein may refer to a computing system that includes at least one processor and memory accessed by the processor.

1 is a block diagram illustrating a query response system 100 and its input / output relationship according to an exemplary embodiment of the present invention. As shown in FIG. 1, the Q & A system 100 may receive a query from a user 20 and provide a response to the query to the user 20. The query response system 100 may be communicatively coupled to the natural language recognition system 200, the meta-knowledge database 300, the pattern template database 400, and the knowledge base 500. 1, at least one of the natural language recognition system 200, the meta-knowledge database 300, the pattern template database 400, and the knowledge base 500 is a query response system 0.0 > 100). ≪ / RTI >

The query response system 100 may receive a natural language query from the user 20. For example, the user 20 can use a terminal connected to the network, can input a natural language query through voice or text to the terminal, and the query response system 100 can receive a natural language query through the network have. The query response system 100 can acquire knowledge (or knowledge data) contained (or stored) in the knowledge base 500 based on natural language queries and generates a response to the natural language query based on the acquired knowledge And can provide a response to the terminal of the user 20 via the network. 1, the Q & A system 100 includes a user interface 110, a query pattern analysis unit 130, a query generation unit 150, a query execution unit 170, and a response generation unit 190 .

The user interface 110 may be connected to a network such as a LAN (Local Area Network) and a WAN (Wide Area Network) or may be connected to a dedicated channel for one-to- have. The user interface 110 can provide the natural language query received from the user 20 to the query pattern analyzing unit 130 and provide the response received from the response generating unit 190 to the user 20.

The query pattern analysis unit 130 may receive the natural language query from the user interface 110 and may provide the natural language query to the natural language recognition system 200. [ The natural language recognition system 200 may generate at least one token list corresponding to the natural language query by analyzing the natural language query (for example, morphological analysis) and provide the generated list to the query pattern analysis unit 130. The token list can include at least one token, and the token can mean a unit of meaningful vocabulary in a natural language query. For example, a token list including tokens of " YiSUN, ", "Hometown "," HALF ", and "?" May be generated from the natural language query " In some embodiments, an investigation such as "on "," on "may be excluded from the token list. The natural language recognition system 200 may also be referred to as a NLU (Natural Language Understanding) system.

In some embodiments, the natural language recognition system 200 may refer to knowledge data contained in the knowledge base 500, as shown in FIG. For example, the natural language recognition system 200 can recognize and extract the token from the natural language query based on the knowledge data stored in the knowledge base 500. [ The natural language recognition system 200 can generate a score for the token list and provide it to the query pattern analyzer 130 together with the token list. The score in the token list may refer to a value indicating the degree to which the lexical token list matches the given natural language query. Examples of the token list provided from the natural language recognition system 200 will be described below with reference to FIG.

The query pattern analyzing unit 130 may generate at least one query pattern by mapping tokens included in the token list received from the natural language recognition system 200 to ontology components. An ontology is a representation of an existing or human-recognizable form in a form that can be handled by a computer. The ontology components can include an entity (or instance), a class, and a property. Additionally, the ontology components may further include relationships (attributes between entities or attributes between classes), function terms, restrictions, rules, events, and the like.

The query pattern analyzing unit 130 can generate a query pattern composed of the ontology components by mapping the tokens extracted from the natural language query to the ontology components. The query pattern may be a combination of ontology components, such as entity (E), class (C), and attribute (P), and the query pattern analysis unit 130 may include meta-knowledge data (Or meta-knowledge) to generate a query pattern. The meta-knowledge data may include information on knowledge data (i.e., as metadata) stored in the knowledge base 500, and information on knowledge data. For example, the meta-knowledge data may include meta information of the entity as described below with reference to FIG. 4 and may include meta information of the attributes as described below with reference to FIGS. 5 and 6 , And may include meta information of the class. In some embodiments, the meta information of the entity may depend on the knowledge data stored in the knowledge base 500, while the meta information of the attribute and / or class may be the data used to build the knowledge base 500 . An example of the query pattern analyzing unit 130 and the meta-knowledge data will be described later with reference to FIG. 2 to FIG.

The query generator 150 can receive the query pattern from the query pattern analyzer 130 and can generate at least one query based on the pattern template (or pattern template data) stored in the pattern template database 400 have. For example, the knowledge base 500 may include knowledge data expressed using an RDF (Resource Description Framework), and the query generation unit 150 may include SPARQL (SPARQL Protocol and RDF Query Language) as an example of an ontology language, You can create a query. The pattern template database 400 may store a plurality of pattern templates, and may provide a pattern template to the query generating unit 150. The pattern template may define a query pattern and its corresponding query template, as described below with reference to FIG. The query generator 150 may generate a query by substituting the values (e.g., URI) of the ontology components included in the query pattern into the query template corresponding to the query pattern received from the query pattern analyzer 130 . An example of the operation of the query generation unit 150 will be described later with reference to FIG.

The query execution unit 170 may provide the query received from the query generation unit 150 to the knowledge base 500 and receive the knowledge data provided by the knowledge base 500 in response to the query. In some embodiments, the query performing unit 170 may receive a plurality of queries corresponding to one natural language query from the query generating unit 150, and may receive knowledge data corresponding to a plurality of queries . The query performing unit 170 may provide the knowledge data received from the knowledge base 500 to the response generating unit 190.

The response generating unit 190 can receive the knowledge data from the query executing unit 170 and generate a response based on the received knowledge data and provide the generated response to the user interface 110. [ In some embodiments, the response generator 190 may receive knowledge data from the query performing unit 170 as a plurality of responses of the knowledge base 500 corresponding to a plurality of queries, Can be determined. In some embodiments, the response generator 190 may generate a natural language response from the knowledge data. In some embodiments, the response generator 190 may generate a natural language query for the natural language query of the user 20 as part of a conversation with the user 20, based on the knowledge data.

An object to which a score is to be awarded, such as a token list, an ontology component, a query pattern, may be referred to as a resource and an evaluation of these resources, i.e., ranking of resources, have. Accordingly, as described below with reference to the figures, the query response system 100 of FIG. 1 can generate a query that meets the intent of a natural language query and, based on pre-built meta-knowledge data, 20 can be quickly generated for the query. Further, since the meta-knowledge data is updated according to the update of the knowledge base, the query generated based on the meta-knowledge data can continuously meet the intention of the natural language query.

Although the query pattern analyzer 130 and the query generator 150 are shown in FIG. 1 as being included in the query response system 100, in some embodiments, as in the natural language recognition system 200, The unit 130 and the query generator 150 may be included in separate independent systems communicatively coupled to the query response system 100 and such independent systems may be referred to as query translation systems.

FIG. 2 is a block diagram illustrating an example of the query pattern analysis unit 130 of FIG. 1 according to an exemplary embodiment of the present disclosure, and FIG. 3 is a block diagram of at least one token list of FIG. 2 4 illustrates an example of entity meta information D310 of FIG. 2 in accordance with an exemplary embodiment of the present disclosure, and FIGS. 5 and 6 illustrate an example of the entity meta information D210 in accordance with the exemplary embodiments of the present disclosure FIG. 7 illustrates examples of at least one query pattern D130 of FIG. 2 in accordance with an exemplary embodiment of the present disclosure; FIG. 7 illustrates examples of attribute meta information D330 of FIG. Hereinafter, Figs. 2 to 7 will be described with reference to Fig.

The query pattern analyzer 130 'of FIG. 2 may receive from the natural language recognition system 200 at least one token list D210 generated from the natural language query, as described above with reference to FIG. 1, - generate at least one query pattern (D130) by mapping the token list to the ontology components with reference to the meta-knowledge data stored in the knowledge database (300 '). 2, the query pattern analyzing unit 130 'may include an entity evaluating unit 131, an attribute evaluating unit 133, a class evaluating unit 135, and a pattern rank unit 137. In addition, the meta-knowledge database 300 'may include entity meta information D310, attribute meta information D330, and class meta information D350.

As shown in FIG. 2, at least one token list D 210 may include a first token list TL ₁ comprising a series of tokens T ₁ , ..., T _n . For example, as shown in the token lists D210 'in FIG. 3, when the natural language query "What is the capital of the country where Barack Obama's father was born?" Is received from the user 20, (TL ₁ ) contains tokens of "Barack Obama", "of", "Father", "A", "Born", "Country", "of", "Capital", " . Similarly, the second to fourth token lists TL ₂ , ..., TL ₄ may each include differently divided tokens, as shown in FIG. The token list may have a score assigned by the natural language recognition system 200, and the score may indicate the degree to which the natural language recognition system 200 in the token list is evaluated as most appropriate. For example, the natural language recognition system 200 may calculate scores for each of the token lists based on predefined sentence structures and / or knowledge data stored in the knowledge base 500. In some embodiments, the query pattern analyzer 130 'may receive scores of each of the token lists along with the token lists. In some embodiments, instead of receiving the score of the token list, the query pattern analyzer 130 'may receive the token lists sorted in high score order from the natural language recognition system 200. [

2, the entity evaluating unit 131, the attribute evaluating unit 133, and the class evaluating unit 135 may classify the meta-knowledge data into at least one (E.g., TL ₁ ) included in the token list D 210 and may evaluate the tokens contained in the received token list. Hereinafter, the operation of each of the entity evaluating unit 131, the attribute evaluating unit 133, and the class evaluating unit 135 will be described.

The entity evaluating unit 131 may evaluate the token corresponding to the entity among the tokens included in the token list by referring to the entity meta information D310 included in the meta-knowledge database 300 '. The entity meta information D310 may include information about the entities included in the knowledge base 500. [ For example, as shown in FIG. 4, the entity meta information D310 'for the entity "Barack Obama" includes a URI (Uniform Resource Identifier), a link count, a Korean name, a source name, an attribute list, And < / RTI > The URI may refer to a unique address for access to the entity "Barack Obama " and may have a value of" 0000148525 ". The link count may refer to the number of links to the entity "Barack Obama ", and may include an in-link to the entity" Barack Obama "and an out-link to the entity" Barack Obama " in-link. < / RTI > An attribute list may contain attributes possessed by the entity " Barack Obama ", and similarly, a class list may include classes that (or belong to) the entity "Barack Obama ". The rank can refer to a value indicating the importance of the entity "Barack Obama ". For example, the rank may be calculated based on the link count, and may be calculated based on the recent query (or query) frequency, etc. for the entity.

In some embodiments, the entity evaluator 131 may calculate the score of the entity corresponding to the token, such that it is proportional to the link count and / or the rank. For example, the entity evaluator 131 calculates the score of the entity (e.g., the average of the positive values) so as to be proportional to the value 770 of the link count and / or the value 790 of the rank for the entity "Barack Obama & can do. In some embodiments, the entity evaluating unit 131 may calculate the score for the entity to a high level if there is a token corresponding to the attribute included in the attribute list among the tokens included in the given token list. Similarly, in some embodiments, the entity evaluator 131 may calculate the score for the entity to a high degree if there is a token corresponding to the class included in the class list among the tokens included in the given token list. In some embodiments, if there is a token corresponding to the attribute / class included in the attribute list / class list among the tokens included in the given token list, the entity evaluating unit 131 may calculate the score for the entity lower Entities can be excluded.

According to at least one of the above-described schemes, the entity evaluator 131 can evaluate the entity and, if a plurality of entities correspond to a particular token, e.g., if a plurality of entities of the same name correspond to a particular token, After calculating the score for each of the entities, the entities can be sorted according to the score.

The attribute evaluation unit 133 may evaluate the token corresponding to the attribute among the tokens included in the token list by referring to the attribute meta information D330 included in the meta-knowledge database 300 '. The property meta information D330 may include information on the attributes included in the knowledge base 500. [ The attribute meta information D330 may include information on a plurality of attributes, and the attributes may have a hierarchical structure depending on the meaning thereof. For example, as shown in FIG. 5, the attributes that the entity may have are included in the attribute meta information D330 ', and the attributes included in the right column in the attribute meta information D330' May be a high-level attribute of the included attribute. The upper level attribute may have a meaning including the meaning of lower level attribute. For example, the attribute " adaptedAs "may correspond to" adapt, "" Invent, "" create, "" invent, "and" create. " Similarly, the attribute " adapter "may have the attribute" creator "as a high-level attribute, and the attribute" adopt " As described above, since the attribute of the high-level has a meaning including the meaning of the property of the low-level, the attribute of the low-level can correspond to a more specific meaning. In the example of FIG. 5, the attribute meta information D330 'is shown to include pairs of attributes of a lower level and attributes of a higher level, but it will be understood that the attributes may have a hierarchical structure such as a tree structure .

In some embodiments, the attribute evaluation unit 133 may determine, based on the hierarchical information of the attributes included in the attribute meta information D330 ', the score of the attribute corresponding to the token so that the attribute of the high- Can be calculated. For example, when two or more attributes correspond to the same token, the attribute evaluation unit 133 can assign a higher score to the lower-level attribute of the attributes, while assigning a lower score to the higher-level attribute . For example, the attribute evaluation unit 133 may calculate the score in inverse proportion to the level from the leaf to the attribute corresponding to the token as a parent in the tree structure of the attributes.

In some embodiments, the attribute meta information D330 of FIG. 2 may include domain information and range information of attributes. A domain may correspond to a subject of an attribute, a scope may correspond to an object or value of an attribute, and each domain and range may be defined as a set of classes. For example, the attribute "write" may include a class "person" as a domain and may include classes "novel "," diary ", & Also, the attribute "bitter" may include the class "food" as a domain and may have the class "boolean" as a range. Figure 6 illustrates examples of domains and ranges of some attributes. Specifically, FIG. 6 shows attributes and ranges according to the classes included in the domain. For example, of the attributes having the class "accident" as a domain, the attribute "agent" may have a class "organization" Similarly, the attribute "awarded " may have the class" award "as a range. In some embodiments, if the token corresponding to the domain or the class of the scope related to the attribute corresponding to the token is included in the token list based on at least one of the domain information and the scope information of the attributes , The score can be calculated such that the score of such an attribute increases.

The class evaluating unit 135 may evaluate the token corresponding to the class among the tokens included in the token list by referring to the class meta information D350 included in the meta-knowledge database 300 '. The class meta information D350 may include information on the classes included in the knowledge base 500. [ The class meta information D350 may include information on a plurality of classes, and similar to the attribute meta information D330, the classes may have a hierarchical structure according to their meaning. The higher-level class may have a meaning including the meaning of the lower-level class, and accordingly, the lower-level class may correspond to a more specific meaning.

In some embodiments, the class evaluation unit 135 calculates the score of the class corresponding to the token so that the higher-level class has a lower score, based on the hierarchical information of the classes included in the class meta information D350 can do. For example, when two or more classes correspond to the same token, the class evaluation unit 135 may assign a higher score to a lower-level class among the classes, . For example, the class evaluating unit 135 may calculate the score to be in inverse proportion to the level from the leaf to the class corresponding to the token in the tree structure of the classes.

In some embodiments, the class meta information D350 of FIG. 2 may include attribute information and range information of the classes. For example, as described above with reference to FIG. 6, a class may have relationships with classes corresponding to attributes and ranges having the class as a domain. 6 may be shared by the attribute evaluating unit 133 and the class evaluating unit 135 and the class evaluating unit 135 may classify at least one of the attributes and range information of the classes A score can be calculated so that the score of such a class is increased if a token corresponding to the class of the attribute or range related to the class corresponding to the token is included in the token list.

As described above, the attribute meta information D330 and the class meta information D350 may include layer information of attributes and layer information of classes, respectively. This hierarchical information may be used to build the knowledge base 500 and may be referred to as an ontology schema. That is, the meta-knowledge data referred to by the query pattern analysis unit 130 'of FIG. 2 includes not only the information produced from the knowledge data included in the knowledge base 500 but also the knowledge base 500 (E.g., D330, D350 in FIG. 2) used to construct the information (e.g.

The pattern rank unit 137 may receive the ontology components and their scores from the entity evaluation unit 131, the attribute evaluation unit 133, and the class evaluation unit 135, and based on this, You can decide. For example, the pattern rank unit 137 can receive scores of entities and entities corresponding to a token from the entity evaluation unit 131, and acquires, from the attribute evaluation unit 133, And receive scores of the classes and classes corresponding to the tokens from the class evaluation unit 135. [ The pattern rank unit 137 can calculate scores for a query pattern that is a combination of entities, attributes, and classes based on the received scores, and calculates at least one query pattern D130 Can be generated.

Referring to FIG. 7, as an example of at least one query pattern D130 in FIG. 2, a plurality of query patterns D130 'may be composed of a token list, a score, and a pattern. As shown in FIG. 7, different scores and different patterns may correspond for the same token list, and tokens corresponding to the examination of the token list in the query pattern may be excluded. For example, as shown in Fig. 7, the query pattern "EPPPC" having the highest score indicates that the tokens "Barack Obama "," Father ", " ), The attribute P, the attribute P, the attribute P, and the class C. Although not shown in FIG. 7, the ontology components included in the query pattern include the corresponding token For example, in the pattern "EPPPC ", the entity" E "may include a URI of" Barack Obama "(e.g., adr: 0000148525) Quot; URI "

FIG. 8 is a diagram illustrating an example of pattern template data included in the pattern template database 400 of FIG. 1 according to an exemplary embodiment of the present disclosure. 1, the query generator 150 of FIG. 1 extracts, from the query pattern received from the query pattern analyzer 130, at least one query based on the pattern template stored in the pattern template database 400, Lt; / RTI > Fig. 8 exemplarily shows pattern template data D410 including a template for generating SPARQL. Hereinafter, FIG. 8 will be described with reference to FIG.

Referring to FIG. 8, the pattern template data D410 may include a plurality of patterns and corresponding templates. As shown in Fig. 8, in the template corresponding to the "EP" pattern (or the query pattern), the URI of the entity can be inserted into "% E%" and the URI of the attribute A name defined in the schema) can be inserted. For example, from the natural language query "What is Barack Obama's father?", An "EP" pattern consisting of the entity "Barack Obama" and the attribute "Father" is generated, and the URI of the entity "Barack Obama" is "adr: 0000148525" , And the URI of the attribute is "adp: father ", the query generation unit 150 refers to the pattern template data D410 of Fig. 8 and determines that the query is defined as" SELECT {adr: 0000148525 adp: father? X. You can create a query. The generated query can be provided to the knowledge base 500 by the query execution unit 170 and the knowledge base 500 can extract a value corresponding to "X" from the knowledge data in response to the query .

In accordance with the above described exemplary embodiments of the present invention described above with reference to the drawings, an example of a process for natural language query "What is the capital of the country where Barack Obama was born?" Will be described below. The results of natural language query analysis by the natural language recognition system 200 can be illustrated as follows.

[

{

"nlp": "Barack Obama",

"[{adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, ...]]}

E {(adr: 0030786216) L [Barack Obama] T [[adc: ability_05295659, adc: cognition_00020729, adc: psychological_feature_00020333, ...]]}

E {(adr: 0000126163) L [Barack Obama] T [[]]} "

  },

  {

    "nlp": "of",

    "log": "L {(J) L [of] T []}"

  },

  {

    "nlp": "father",

    "father" T []} Father C {(adc: fatherfigure_09435082) L [fatherfigure] T []}

E {(adr: 0031245687) L [Father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, adc: movie_06205452, ...]]}

E {(adr: 0031545743) L [father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, adc: communication_00028764, ...]]}

E {(adr: 0031108925) L [Father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, adc: communication_00028764, ...]]}

E {(adr: 0031334934) L [Father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, adc: movie_06205452, ...]]}

E {(adr: 0012072422) L [father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, adc: movie_06205452, ...]]}

E {(adr: 0000210970) L [Father] [[adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, adc: organism_00003226, ...]]}

E {(adr: 0030567899) L [Father] [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, ...]]}

E {(adr: 0000499795) L [Myeongsoo Park] [[[adc: person_00006026, adc: institution_07563940, adc: entity_00001740, adc: group_00026769, ...]]}

E {(adr: 0000112273) L [Theodore Iceke] T [[adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, ...]]}

E {(adr: 0000424316) L [Park Dong-soo] T [[adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, ...]]}

W {(WIKI) L [Father] T [[]]} Father U {(U_NE) L [AF] T []}

  },

  {

    "nlp": "

    "log": "L {(J) L [] T []}"

  },

  {

    "nlp": "born",

    L {[birthday] L [undefined] T []} P {(adp: bornOn) L [ "

  },

  {

    "nlp": "country",

    [country] T []} country C {(adc: state_07673557) L [body politic] T []}

E {(adr: 0000347468) L [country] T [[adc: person_00006026, adc: entity_00001740, adc: actor_09145973, adc: performer_09740423, ...]]}

E {(adr: 0000501329) L [country] T [[adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, ...]]}

E {(adr: 0000478478) L [Nara City] T [[adc: city_08005407, adc: geographical_area_08050136, adc: state_capital_08163309, ...]]}

E {(adr: 0000247685) L [Country] T [[adc: cognition_00020729, adc: abstraction_00020486, adc: content_05473476, ...]]}

E {(adr: 0000620719) L [[adc: entity_00001740, adc: sp_GeoEntity, adc: facility_03194800, adc: structure_04174544, ...]]}

E {(adr: 0000078432) L [white country] T [[adc: person_00006026, adc: entity_00001740, adc: causal_agent_00005598, ...]]}

W {(WIKI) L [country] T [[]]} country U {(U_NE) L [CH] T []}

  },

  {

    "nlp": "of",

    "log": "L {(J) L [of] T []}"

  },

  {

    "nlp": "Capital",

    T []} P {(adp: capitalOf) L [Capital] T []} Capital

C {(adc: national_capital_08159919) L [national capital] T []} capital C {(adc: plumbing_fixture_03819002) L [plumbing fixture] T []}

E {(adr: 0000340502) L [Capital] T [[adc: cognition_00020729, adc: content_05473476, adc: concept_05498421, ...]]}

E {(adr: 0000510352) L [Capital] T [[adc: abstraction_00020486, adc: term_05916288]]}

E {(adr: 0000652025) L [Capital] T [[adc: entity_00001740, adc: sp_GeoEntity, adc: land_08750552, adc: object_00016236, adc: island_08734830]]}

E {(adr: 0000048066) L [Capital] T [[]]}

E {(adr: 0000610209) L [Capital] T [[adc: entity_00001740, adc: sp_GeoEntity, adc: land_08750552, adc: object_00016236, adc: island_08734830]]}

E {(adr: 0031955789) L [Capital] T [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, ...]]}

E {(adr: 0000387233) L [Capital] T [[adc: entity_00001740, adc: sp_GeoEntity, adc: land_08750552, adc: object_00016236, adc: island_08734830]]}

E {(adr: 0031979581) L [Capital] T [[adc: abstraction_00020486, adc: event_00025950, adc: social_event_06841042, ...]]}

E {(adr: 0000531168) L [degree] T [[adc: abstraction_00020486, adc: term_05916288]]} W {

  },

  {

    "nlp": "",

    "log": "L {(J) L [] T []}"

  },

  {

    "nlp": "?",

    "log": "L {(PUNCT) L [?] T []}"

  }

]

From the token analysis result, a query can be generated by the query pattern analysis unit 130 and the query generation unit 150, and the SPARQL can be generated as follows.

SELECT DISTINCT? X

        {

            adr: 0000148525 adp: father? V1.

            ? V1 adp: bornIn? B.

            ? b a adc: state_07673557.

            ? b adp: capital? X.

        }

The response generator 190 can receive the response of the knowledge base 500 for the above query (i.e., "Nairobi" as a value corresponding to X) and responds to the natural language query and the received response as follows Can be generated.

"The capital of the country where Barack Obama's father was born is Nairobi."

9 is a flow diagram illustrating a method for generating a query for a knowledge base in accordance with an exemplary embodiment of the present disclosure. In some embodiments, FIG. 9 may be performed by the query response system 100 of FIG. 1, and hereinafter, FIG. 9 will be described with reference to FIG.

In step S20, an operation of obtaining the tokens extracted from the natural language query can be performed. For example, the query pattern analysis unit 130 of the QMS system 100 may include at least one token list from the natural language recognition system 200 and may obtain the tokens included in the token list.

In step S40, an operation of generating a query pattern by mapping the tokens to the ontology components may be performed with reference to the meta-knowledge data D300. For example, the query pattern analyzer 130 of the query response system 100 may refer to the meta-knowledge data D300 stored in the meta-knowledge database 300 and may compare tokens in the token list with ontology components , For example, entity E, attribute P, class C, or the like. The query pattern analysis unit 130 may calculate a score for a query pattern corresponding to the token list based on the meta-knowledge data D300, and the score for the query pattern corresponds to the tokens included in the token list Can be calculated from the scores of each of the ontology components. The query pattern analyzing unit 130 'can calculate the score of the query pattern by summing the scores of the respective ontology components and sort the plurality of query patterns according to the scores. An example of step S20 will be described later with reference to Fig.

In step S60, an operation of generating a query from the query pattern may be performed with reference to the query pattern template D400. For example, the query generator 150 of the Q & A 100 can refer to a query pattern template D400 stored in the pattern template database 400, and generates a URI corresponding to the token in the template corresponding to the query pattern To generate a query.

10 is a flow chart illustrating an example of step S40 of FIG. 9 in accordance with an exemplary embodiment of the present disclosure. As described above with reference to FIG. 9, an operation of generating a query pattern by mapping the tokens to the ontology components by referring to the meta-knowledge data D300 'in step S40' of FIG. 9 may be performed. 10, the step S40 'may include a plurality of steps S42, S44, S46, and S48, and the meta-knowledge data D300' may include entity meta information D310 ' Information D330 ' and class meta information D350 '. Although three steps S42, S44, and S46 are shown in FIG. 10 as being performed sequentially, in some embodiments, the three steps S42, S44, and S46 may be performed in parallel.

In step S42, an operation of calculating the score of the entity with reference to the entity meta information D310 'may be performed. For example, the score for the entity corresponding to the token included in the token list may be calculated based on the importance possessed by the entity, or may be calculated based on the attribute included in the token list or the relationship between the class and the entity have.

In step S44, an operation of calculating the score of the attribute by referring to the attribute meta information D330 'may be performed. For example, the score for the attribute corresponding to the token included in the token list may be calculated so that the low-level attribute has a high score based on the hierarchical information of the attribute, or based on the attribute domain and range information .

In step S46, an operation of calculating the score of the class can be performed by referring to the class meta information D350 '. For example, the score for the class corresponding to the token included in the token list may be calculated so that the lower-level class has a higher score based on the hierarchical information of the class, and the attribute having the class as a domain and the May be calculated based on the range information.

In step S48, an operation of calculating the score of the query pattern may be performed. For example, the score of the query pattern can be calculated based on the scores calculated in the three steps (S42, S44, S46). Accordingly, the query pattern evaluated as best suited to a natural language query can have the highest score. A plurality of query patterns may be generated for a plurality of token lists and a plurality of query patterns may be generated for the same token list.

11 is a block diagram illustrating a meta-knowledge building system 700 and its input / output relationship in accordance with an exemplary embodiment of the present disclosure. As shown in FIG. 11, the meta-knowledge building system 700 may be communicatively coupled to the meta-knowledge database 300 and the knowledge base 500. As described above with reference to FIG. 1, knowledge base 500 may include knowledge data, and meta-knowledge database 300 may include meta-knowledge data.

The knowledge base construction system 600 can update the knowledge base 500. [ For example, the knowledge base construction system 600 can be controlled by users made up of experts and can add new knowledge data to the knowledge base 500 or modify knowledge data stored in the knowledge base 500 . In addition, the knowledge base construction system 600 can be connected to a network such as the Internet, and can update the knowledge base 500 based on information collected through the network. 2, the meta-knowledge database 300 may include data used to build the knowledge base 500 (e.g., D330, D350 of FIG. 2) 600 may update the knowledge base 500 with reference to the information. For example, the knowledge base construction system 600 may generate triples constituting knowledge data based on hierarchical information of attributes and / or hierarchical information of classes included in the meta-knowledge database 300 .

The meta-knowledge building system 700 may update the meta-knowledge database 300 based on the knowledge data stored in the knowledge base 500. For example, when the knowledge base 500 is updated by the knowledge base building system 600, the meta-knowledge building system 700 may generate meta-knowledge data from the added knowledge data or modified knowledge data Knowledge data may be added to the meta-knowledge database 300 or meta-knowledge data stored in the meta-knowledge database 300 based on the generated meta-knowledge data. Accordingly, the meta-knowledge database 300 may reflect the updated knowledge base 500 and may be used in the processing of the natural language query to generate appropriate query patterns by a query response system (e.g., 100 in FIG. 1) .

As described above, exemplary embodiments have been disclosed in the drawings and specification. While the embodiments have been described herein with reference to specific terms, it should be understood that they have been used only for purposes of describing the technical idea of the invention and not for limiting the scope of the invention as defined in the claims . Therefore, those skilled in the art will appreciate that various modifications and equivalent embodiments are possible without departing from the scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (10)

A query pattern analyzer configured to generate at least one query pattern sorted according to a ranking by mapping tokens extracted from a natural language query to ontology components based on meta-knowledge data; And
And a query generator configured to generate at least one query for the knowledge base from the at least one query pattern based on the query pattern template,
Wherein the meta-knowledge data includes meta information of the ontology components in the knowledge base,
Wherein the meta-knowledge data includes meta information of an attribute including at least one of hierarchical information of attributes, domain information, and range information,
Wherein the query pattern analyzing unit includes an attribute evaluation unit configured to calculate a score of the at least one attribute based on meta information of at least one attribute corresponding to the token. The method according to claim 1,
Wherein the meta-knowledge data includes meta information of an entity including at least one of an importance of an entity, a link count, an attribute list, and a class list,
Wherein the query pattern analyzing unit includes an entity evaluating unit configured to calculate a score of the at least one entity based on meta information of at least one entity corresponding to the token. The method of claim 2,
Wherein the entity evaluator calculates a score of the at least one entity to be proportional to at least one of the importance and the link count. The method of claim 2,
Wherein the entity evaluating unit calculates a score of the at least one entity so as to increase when the attribute included in the attribute list or the token corresponding to the class included in the class list among the extracted tokens Q & A system. delete The method according to claim 1,
Wherein the attribute evaluation unit calculates the score of the at least one attribute such that the attribute of the higher level has a lower score based on the hierarchical information of the attributes. The method according to claim 1,
Wherein the attribute evaluating unit is configured to increase a token corresponding to a class of a related domain or range of the at least one attribute of the extracted tokens based on at least one of domain information and range information of the attributes, And the score of one attribute is calculated. The method according to claim 1,
Wherein the meta-knowledge data includes meta information of a class including hierarchical information of classes,
Wherein the query pattern analyzing unit includes a class evaluating unit configured to calculate a score of the at least one class based on meta information of at least one class corresponding to the token. The method of claim 8,
Wherein the class evaluation unit calculates the score of the at least one class so that the higher-level class has a lower score, based on the hierarchical information of the classes. The method according to claim 1,
Wherein the query pattern analyzer comprises a query pattern rank portion configured to determine a ranking of the at least one query pattern based on scores of ontology components included in the at least one query pattern. .

Images