KR101987915B1 - System for generating template used to generate query to knowledge base from natural language question and question answering system including the same - Google Patents

Abstract

The present invention relates to a system for generating a template used for generating a query about a knowledge base from a natural language query, and a query response system including the same. According to one exemplary embodiment of the present invention, the system comprises: a natural language query pretreatment unit transferring a natural language query to a natural language recognition system to acquire resources including each of ontology components corresponding to tokens extracted from the natural language query; a graph analysis unit generating graph data including resources based on an ontology schema of the knowledge base formed with knowledge data units, and generating at least one knowledge data unit from the graph data; and a template generation unit generating a sample query from the at least one knowledge data unit and generating a template corresponding to a pattern of the resources from the sample query.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for generating a template used for generating a query for a knowledge base from a natural language query, and a query response system including the same. 2. Description of the Related Art Generally,

TECHNICAL FIELD The present invention relates to a knowledge base, and more particularly, to a system for generating a template used for generating a query from a natural language query to a knowledge base, and a query response system including the same.

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 template used to generate a query for a knowledge base from a natural language query.

According to an aspect of the present invention, there is provided a system for generating a template for generating a query from a natural language query to a knowledge base, the system comprising: A natural language query preprocessing unit configured to acquire resources each including resources corresponding to the ontology components of the knowledge base unit, graph data including resources based on the ontology schema of the knowledge base composed of knowledge data units A graphical analysis unit configured to generate at least one knowledge data unit from the graph data, and a template generation unit configured to generate a sample query from at least one knowledge data unit and generate a template corresponding to the pattern of resources from the sample query Could have.

According to an exemplary embodiment of the present invention, the natural language query preprocessor may be configured to collect a plurality of natural language queries through a network or to obtain user natural language queries provided to the query response system.

According to an exemplary embodiment of the present invention, the natural language query preprocessing unit can be configured to obtain a natural language query that is not matched to the template in the processing by the query response system among the natural language queries provided to the query response system.

According to an exemplary embodiment of the present invention, the graph analyzing unit generates a node from a resource including an entity or a class among ontology components and generates an edge from a resource including an attribute and a relation among the ontology components, A resource filtering unit configured to filter resources by selectively removing each of the nodes and edges included in the graph data, and at least one knowledge data unit from the graph data including the filtered resources, And a knowledge data unit generation unit configured to generate the knowledge data unit.

According to an exemplary embodiment of the present invention, the resource filtering unit may be configured to filter resources by removing at least one of edges having no edge and nodes not connected to a node.

According to an exemplary embodiment of the present invention, each of the resources may further include a weight corresponding to the degree of matching of the token and the ontology component, and the resource filtering unit may include a predefined And filtering the resources by removing interconnected nodes and edges having aggregated weights below the threshold.

According to an exemplary embodiment of the present invention, the template generation unit may include a query verification unit configured to verify a sample query based on knowledge base knowledge data.

According to an exemplary embodiment of the present invention, the query verification unit may further include an administrator interface. The query verification unit may provide a natural language query and a sample query to an administrator through an administrator interface, ) And information about the change.

According to an exemplary embodiment of the present invention, the template generator may be configured to store the template generated by the template store in a template repository accessed by the query response system.

According to an exemplary embodiment of the present invention, the knowledge data unit may be triple and the query may be SPARQL (SPARQL Protocol and RDF Query Language).

According to the system and method according to the technical idea of the present invention, a verified template used for generating a query from a natural language query can be automatically generated.

In addition, according to the system and method according to the technical idea of the present invention, a query response system can be easily implemented due to automatically generated templates.

In addition, according to the system and method according to the technical idea of the present invention, a query response system capable of quickly responding to various natural language queries due to templates generated beforehand from natural language queries can be implemented.

The effects obtainable in the exemplary embodiments of the present invention are not limited to the above-mentioned effects, and other effects not mentioned can be obtained from the following description based on conventional knowledge in the art to which the exemplary embodiments of the present invention belong Can be clearly understood and understood. That is, unintended effects of implementing the exemplary embodiments of the present invention may also be derived from those of ordinary skill in the art from the exemplary embodiments of the present invention.

1 is a block diagram illustrating a template generation 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 a template generation system of Figure 1 and its input / output relationship in accordance with an exemplary embodiment of the present invention.
Figure 3 is a block diagram illustrating an example of the graph analyzer of Figure 2 in accordance with an exemplary embodiment of the present invention.
Figures 4A and 4B are illustrations of examples of an ontology schema in accordance with an exemplary embodiment of the present invention.
5 is a diagram showing an example of graph data according to an exemplary embodiment of the present invention.
6 is a diagram illustrating an example of a template according to an exemplary embodiment of the present invention.
Figure 7 is a block diagram illustrating an example of the template generator of Figure 2 in accordance with an exemplary embodiment of the present invention.
8 is a flow diagram illustrating a method for generating a template for use in generating a query for a knowledge base from a natural language query in accordance with an exemplary embodiment of the present invention.

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

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

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries 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 template generation system 100 and its input / output relationship according to an exemplary embodiment of the present invention. 1, the template generation system 100 may be communicatively coupled to the natural language recognition system 500, the knowledge base 700, and the template database 900. The query response system 300 can also receive queries from the user 20 such as natural language queries and communicate with the user 20 by communicating with the natural language recognition system 500, knowledge base 700 and template database 900. [ Lt; RTI ID = 0.0 > a < / RTI > The query response system 300 may generate a query for the knowledge base 700 from a natural language query with reference to the templates stored in the template database 900 and the template generation system 100 may generate a query for the knowledge base 700, The template database 900 may generate templates for the template database 300.

The blocks 100, 300, 500, 700, and 900 shown in FIG. 1 may communicate with each other through a network or may communicate with each other through a dedicated channel for one-to-one communication. In addition, two or more of the blocks 100, 300, 500, 700, 900 shown in FIG. 1 may be included in one system (e.g., a computing system) May be included in the Q & A system 300.

The query response system 300 may receive the natural language query from the user 20. For example, the user 20 can use a terminal connected to the network, input natural language queries through voice or text to the terminal, and the query response system 300 can receive natural language queries through the network have. The query response system 300 can transmit the natural language query to the natural language recognition system 500 and receive the natural language query formual data from the natural language recognition system 500. [ As will be described later, the formal data generated by the natural language recognition system 500 with respect to the natural language query can be used by the template generation system 100 as well as the query response system 300. [

The natural language recognition system 500 may be referred to as a NLU (Natural Language Understanding) system and may generate regular data comprised of resources, each containing ontology components corresponding to tokens included in a natural language query have. An ontology consists of an entity (E) (or instance), a class (class C), a property (property P), and an ontology element in the form of a computer- , And a value (value V) (or literal L). Additionally, the ontology components may further include relationships (attributes between entities or attributes between classes), function terms, restrictions, rules, events, and the like.

The natural language recognition system 500 can generate a list of tokens containing tokens extracted from the natural language query and generate resources corresponding to the tokens included in the token list. For example, for a natural query of "Barack Obama's wife's age?", The natural language recognition system 500 extracts the tokens "Barack Obama", "of", "wife", " Age, "" and "? &Quot;. The natural language recognition system 500 creates a resource for the token "Barack Obama" by mapping the token "Barack Obama" to the entity "E {adr: 0000148525, 4.18105395E9}" with reference to the knowledge data stored in the knowledge base 700 can do. Similarly, the natural language recognition system 500 may generate a resource for the token "age" by mapping the "age" to the attribute "P {adp: age, 3418252.2} ". The natural language recognition system 500 may map to the attribute "P {adp: spouse, 5699.2476}" corresponding to the "spouse" for the token "wife " according to the ontology schema of Fig. 4B, for example.

In the examples of the entities and attributes described above, the first item in the braces may represent a Uniform Resource Identifier (URI) of the entity and attribute, and the second item may be a weight of the entity and attribute (or resource) . In addition, in some embodiments, the natural language recognition system 500 may exclude instances such as "on" and "on " In this specification, a resource may be referred to as a unit including a single component of the ontology as described above, or may refer to a set of resources corresponding to a plurality of tokens, such as a token list generated from a natural language query.

Different resources may be created for the same token list, and resources may be created for each of the token lists containing different tokens. The natural language recognition system 500 can refer to the knowledge data stored in the knowledge base 700 in the process of mapping the token included in the natural language query to the ontology component and maps the token to the ontology component defined in the knowledge data Resources can be created.

In some embodiments, the natural language recognition system 500 may weight resources. For example, a plurality of ontology components may correspond to a single token, and the natural language recognition system 500 may determine the importance (e.g., link count of the entity) of the ontology component for each of the plurality of ontology components and / RTI > and / or relationships with other ontology components. ≪ RTI ID = 0.0 > Accordingly, the weight may have a value indicating the degree of matching between the token and the ontology component in the natural language query. The weights of the resources may be used by the query response system 300 to determine the ranking of resources and may be used to filter resources in the template generation system 100, as described below.

The query response system 300 can search the template stored in the template database 900 based on the formatted data and generate the query according to the template corresponding to the formatted data. In some embodiments, the knowledge base 700 may include knowledge data expressed using a Resource Description Framework (RDF), and a triple as the knowledge data unit may be used. Accordingly, the query response system 300 can generate a SPARQL (SPARQL Protocol and RDF Query Language) query as a query to the knowledge base 700. Hereinafter, exemplary embodiments of the present invention will be described mainly with reference to SPARQL, but it will be understood that the exemplary embodiments of the present invention are not limited thereto. The query response system 300 can obtain the knowledge (or knowledge data) contained (or stored) in the knowledge base 700 by providing the generated query to the knowledge base 700, and based on the acquired knowledge A response to a natural language query can be generated. Finally, the Q & A system 300 can provide a response to the terminal of the user 20.

The template that the Q & A system 300 refers to when generating a query may be composed of a pattern of resources of the fixed data provided by the natural language recognition system 500 and a corresponding SPARQL template, as illustrated in FIG. 6, The template database 900 may store a plurality of templates. Accordingly, as the amount of the template stored in the template database 900 increases, the natural language query processable by the query response system 300 can be increased, while the accuracy of the response to the natural language query depends on the accuracy of the template .

The template generation system 100 can acquire a natural language query, generate a template from a natural language query, and store the generated template in the template database 900. The template generation system 100 can generate a valid template by using the natural language query as an input for generating a template, which is the same as that which can be provided by the user 20 to the query response system 300. [ The template generation system 100 generates resources for providing to the natural language recognition system 500 in the course of generating a template from a natural language query, an ontology schema used for constructing the knowledge base 700, knowledge Data and the like can be used.

As described below with reference to the drawings, the template generation system 100 may automatically generate a validated template by verifying the template used by the query response system 300 to generate a query from a natural language query. In addition, the template generation system 100 may facilitate the implementation of the query response system 300 by supplying templates to the template database 900. In addition, a query response system 300 capable of quickly responding to various natural language queries due to templates generated in advance by the template generation system 100 can be implemented. Examples of the template generation system 100 will now be described with reference to the figures.

2 is a block diagram illustrating an example of the template generation system 100 of FIG. 1 and its input / output relationship in accordance with an exemplary embodiment of the present invention. As described above with reference to FIG. 1, the template generation system 100 'of FIG. 2 may generate a template from a natural language query. 2, the template generating system 100 'may include a query preprocessing unit 120, a graph analyzing unit 140, and a template generating unit 160. Hereinafter, Will be described with reference to FIG.

The query preprocessing unit 120 can obtain the natural language query and obtain the natural language query, that is, the resources from the natural language recognition system 500 'by providing the natural language query to the natural language recognition system 500' . The query preprocessing unit 120 can acquire natural language queries in various ways. In some embodiments, the query preprocessor 120 may collect queries in natural language through a network, such as the Internet. In some embodiments, the query preprocessing unit 120 may receive a natural language query as a sample for template generation from an administrator, as described below with reference to FIG. In some embodiments, the query preprocessing unit 120 may enable the user 20 to query the query response system 300 via an interface (not shown) included in the template generation system 100 ' The natural language query provided to the user 300 can be obtained. In some embodiments, the query preprocessing unit 120 may obtain a natural language query that does not have a matching template among the natural language queries provided by the user 20 through communication with the query response system 300. The pre-processing unit 120 can provide a natural language query to the natural language recognition system 500 and receive the natural language query data from the natural language recognition system 500. The query preprocessing unit 120 may provide the graph data analysis unit 140 with the format data for the natural language query.

The graph analyzing unit 140 may generate graph data from the form data received from the query preprocessing unit 120 and may generate at least one knowledge data unit from the graph data. The graph analyzing unit 140 may generate graph data based on an ontology schema D20 of the knowledge base 700. [ The ontology schema D20 is data used in building the knowledge base 700, and defines the ontology components and their relationships, as described below with reference to Figs. 4A and 4B. The graph data may define a graph that includes the resources contained in the structured data as nodes and / or edges.

The graph analyzer 140 may filter the resources included in the graph data and generate at least one knowledge data unit in the filtered graph. The knowledge data unit is a data unit of the knowledge base 700 and may be triple. A triple can include a subject, a predicate, and an object, and subjects, predicates, and objects of a triple can correspond to ontology components, such as entities, attributes, and classes. Knowledge base 700 may store a plurality of triples corresponding to knowledge data and may result in an ontology component included in the triple by searching for triples in response to a SPARQL query. The graph analyzer 140 may provide at least one triple to the template generator 160. An example of the operation of the graph analyzing unit 140 will be described later with reference to FIGS.

The template generating unit 160 may receive at least one triple from the graph analyzing unit 140 and may generate a sample query from at least one triple. For example, the template generation unit 160 may generate a sample query based on the SPARQL syntax. The sample query may include an ontology component included in the triple received from graph analyzer 140. [ Further, in some embodiments, the template generation unit 160 may verify the sample query, as described below with reference to FIG.

The template generating unit 160 may generate a template from the sample query. For example, the template generation unit 160 may generate a template corresponding to a pattern of resources provided from the query preprocessing unit 120. [ As will be described later with reference to FIG. 6, the template generating unit 160 may generate a template by replacing the ontology component having a value corresponding to the resource in the sample query with a variable having the type of the ontology component. The template generating unit 160 may store the template in the template database 900 '.

FIG. 3 is a block diagram illustrating an example of the graph analyzer 140 of FIG. 2 in accordance with an exemplary embodiment of the present invention, and FIGS. 4A and 4B illustrate examples of an ontology schema according to an exemplary embodiment of the present invention. And FIG. 5 is a diagram illustrating an example of graph data according to an exemplary embodiment of the present invention.

Referring to FIG. 3, the graph analyzing unit 140 'of FIG. 3 may receive resources as formatted data from the query preprocessing unit 120 of FIG. 2, as described above with reference to FIG. 2, I.e., at least one triple, to the template generating unit 160 of FIG. The graph analyzing unit 140 'may include a graph generating unit 142, a resource filtering unit 144 and a knowledge data unit generating unit 146 and may refer to the ontology schema D20.

The graph generating unit 142 may generate graph data including resources as nodes and / or edges by referring to the ontology schema D20. As described above, the ontology schema D20 is data used to build the knowledge base 700 of FIG. 1, and may define the relationships of the ontology components and their relationships. For example, FIG. 4A shows an example of an ontology schema D20 that defines the relationship between ontology components, and FIG. 4B shows an example of an ontology schema D20 that defines hierarchical information between homogeneous ontology components .

Referring to FIG. 4A, an ontology schema D20 may define a relationship between ontology components. That is, the class "person" may have a number of attributes such as "name "," birthday ", etc., and each of the attributes may be associated with an ontology component, e.g., class, value. Accordingly, entities belonging to the class "person ", such as entity" Barack Obama ", may have properties such as "name "," birthday ", "birthplace ", & The knowledge base 700 of FIG. 1 built on the ontology schema of FIG. 4A may have knowledge data units, or triples, that define other entities and / or values corresponding to the attributes of entity "Barack Obama" . Knowledge base 700 may be continuously updated and may acquire and store entities or values in a variety of ways, for example, where there is no entity or value corresponding to an attribute of the entity.

Referring to FIG. 4B, the ontology schema D20 can define hierarchical information between ontology elements of the same type. The ontology components, such as classes, attributes, etc., may have a hierarchical structure depending on their meaning. As shown in FIG. 4B, the attribute / class included in the right column may be a property / class of the attribute / class of the upper level included in the left column. A higher level attribute / class may have a meaning that includes the meaning of a lower level attribute / class. For example, the attribute "husband" may have the property "spouse" as a property of a higher level, and the property "wife" Similarly, the class "President" can have the class "politician" as a high-level class.

Referring to FIG. 5, the graph generating unit 142 of FIG. 3 can generate graph data from resources generated from the natural language query "Barack Obama's wife's age?", FIG. Of the graph G50. The graph generating unit 142 may generate a node N51 corresponding to the entity "Barack Obama ". The entity "Barack Obama" can belong to the class "person ", and the graph generation unit 142 refers to the ontology schema of FIG. 4A to determine the attributes of the entity" Barack Obama " &Quot;, "spouse ", and so on. As shown in FIG. 5, the edge corresponding to the attribute "spouse" may be a bi-directional edge.

The graph generating unit 142 may generate the nodes N52, N53, N54 and N55 connected to the other end of the edges by referring to the ontology schema of FIG. 4A, and generate the generated nodes N52, N53, N54, N55) can have a type depending on the attribute. For example, as shown in Fig. 5, the node N54 connected to the attribute "age " may correspond to the value" number ", and the node N53 connected to the attribute " , And the node N54 connected to the attribute "birthday" may correspond to the value "date ". Further, the node N55 connected to the attribute "spouse" may correspond to the class "person ". The graph generating unit 142 may additionally consider the attributes for the node N55 connected by the attribute "spouse" in accordance with the attribute "spouse" generated from the token "wife " Accordingly, although not shown, a plurality of edges corresponding to a plurality of attributes can be generated for the node N55 similarly to the node N51.

Referring again to FIG. 3, the resource filtering unit 144 may receive the graph data from the graph generating unit 142 and may filter the resources in the graph data. In some embodiments, the resource filtering unit 144 may filter the resources based on the structure of the graph data. For example, the resource filtering unit 144 may remove resources that form an incomplete graph, such as nodes that do not have edges, edges that do not connect to nodes, and so on. Further, in some embodiments, the resource filtering unit 144 may remove a group having a relatively small number of nodes when there are two or more groups of nodes that are not interconnected by edges. In some embodiments, the resource filtering unit 144 may filter the resources based on the weight of the resource. The resource filtering unit 144 may remove, in the graph data, interconnected nodes and edges having aggregated weights below a predefined reference value. As described above with reference to FIG. 1, the natural language recognition system 500 can provide, as a resource, a weight indicating the degree of matching of the token and the ontology component together with the ontology component corresponding to the token, 144 may filter the resources in the graph data based on the weights included in the resources. 5, the graph generating unit 142 removes some of the nodes N52, N53, N54 from the graph G50, and thus the nodes N51, N55, N56, and N56, Gt; N51, < / RTI > N55, N56. The resource filtering unit 144 may provide the knowledge data unit generation unit 146 with graph data including the filtered resources.

The knowledge data unit generation unit 146 may receive the graph data including the filtered resources from the resource filtering unit 144 and may generate at least one knowledge data unit from the received graph data. For example, the knowledge data unit generation unit 146 can receive graph data including edges connected to the shaded nodes N51, N55, and N56 and nodes N51, N55, and N56 in FIG. 5 And can generate two knowledge data units from the graph data. That is, two triples can be created: "E: Barack Obama? Spouse? C: Person" and "C: Person? The triple generated by the knowledge data unit generation unit 146 may be provided to the template generation unit 160 of FIG.

6 is a diagram illustrating an example of a template according to an exemplary embodiment of the present invention. Specifically, Figure 6 shows a plurality of patterns of ontology components and a corresponding plurality of SPARQL templates.

2, the template generation unit 160 generates a sample query from at least one knowledge data unit received from the graph analysis unit 140 (or the knowledge data unit generation unit 146 in FIG. 3) Can be generated. For example, according to the example of FIG. 5, when the template generating unit 160 receives two triples "E: Barack Obama? Spouse? C: Person" and "C: Person? , The template generating unit 160 can generate the following SPARQL query as a sample query.

SELECT? X {

adr: 0000148525 adp: spouse? Y

? Y adp: age? X

}

In the above sample query, "adr: 0000148525" represents the URI of the entity "Barack Obama", "adp: spouse" represents the URI of the property "spouse", and "adp: age" represents the URI of the property "age" . Accordingly, an entity "Y " having an attribute " Barack Obama " and an attribute" Spouse "can be retrieved and a value" X "

The template generating unit 160 may generate a template from the sample query. For example, the tokens extracted from "Barack Obama's wife's age?" Consist of the entity "Barack Obama", the attribute "spouse" and the attribute "age" , The pattern of the ontology constituent elements of the natural language query can be expressed as "EP1P2 ". The sample query of the above-described example can be generated from the natural language query "What is Barack Obama's wife? &Quot;, and the template generation unit 160 converts the ontology component having the value corresponding to the resource in the sample query, To create a template. When a portion starting with "%" and ending with "%" is defined as a variable, a SPARQL template corresponding to the pattern "EP1P2" can be generated as shown in FIG. The templates shown in FIG. 6 may be stored in the template database 900 of FIG. 1 and may be referenced by a query response system 300 having a pattern of ontology components corresponding to natural language queries.

FIG. 7 is a block diagram illustrating an example of the template generation unit 160 of FIG. 2 according to an exemplary embodiment of the present invention. Specifically, the template generating unit 160 'of FIG. 7 can verify the sample query and generate a template from the verified sample query. 7, the template generating unit 160 'may include a query verifying unit 162. [

In some embodiments, the query verifier 162 may verify the sample query based on the knowledge data stored in the knowledge base 700 '. That is, the query verifier 162 may provide a sample query to the knowledge base 700 ', and may verify the sample query based on the results provided by the knowledge base 700' in response to the sample query. For example, if an error occurs in the knowledge base 700 'for a sample query, the sample query may fail to validate, whereas if the knowledge base 700' provides a normal response for the sample query, Can be successfully verified.

In some embodiments, the query verifier 162 may be verified by the administrator 30. To this end, as shown in FIG. 7, the template generation system may further include an administrator interface 180. [ The manager interface 180 can provide the manager 30 with the sample query received from the query verifier 162 or the template generated from the sample query and receives the feedback on the sample query from the manager 30 And provide it to the query verification unit 162. For example, the manager interface 180 may receive information about the suitability and / or modification of the sample query from the manager 30 and may provide the received information to the query verifier 162 . The query verifier 162 can determine whether the verification of the sample query is successful or modify the sample query according to the information on the change of the sample query according to the information about the suitability of the sample query. In some embodiments, the manager interface 180 may receive a natural language query (or sample natural language query) for template generation from the manager 30.

8 is a flow diagram illustrating a method for generating a template for use in generating a query for a knowledge base from a natural language query in accordance with an exemplary embodiment of the present invention. 8, the method of FIG. 8 may include a plurality of steps S10, S30, S50, S70, S90, and an ontology schema D20 may be used. In some embodiments, the method of FIG. 8 may be performed by the template generation system 100 'of FIG. 2, and hereinafter, FIG. 8 will be described with reference to FIG.

In step S10, an operation of obtaining the resources generated from the natural language query can be performed. For example, the query preprocessing unit 120 can acquire natural language queries in various ways and can acquire resources from the natural language recognition system 500 'by providing the acquired natural language queries to the natural language recognition system 500' have. The resource may include an ontology component (e.g., entity, attribute, class, etc.) corresponding to the token included in the natural language query and may further include a weight indicating a degree of matching between the token and the ontology component.

In step S30, an operation of generating graph data by referring to the ontology schema D20 may be performed. The ontology schema D20 is data used to build (or update) a knowledge base (e.g., 700 in FIG. 1) referenced by a query response system (e.g., 300 in FIG. 1) Can be defined. For example, the graph analyzer 140 may generate graph data including ontology elements included in resources as nodes or edges by referring to the ontology schema D20. In addition, the graph analyzer 140 may filter resources based on the structure of the graph data and / or the weights of nodes and edges in the graph data, and the resources may generate filtered graph data.

In step S50, an operation of generating a knowledge data unit may be performed. For example, the knowledge data unit may be triple, and the graph analysis unit 140 may generate at least one triple in the graph data in which the resources are filtered. For example, a triple can be created from two nodes and an edge connecting two nodes.

In step S70, an operation of generating a sample query can be performed. For example, the template generating unit 160 may generate a sample query from at least one triple received from the graph analyzing unit 140. The sample query can include the URI of the ontology component contained in the triple. In some embodiments, the sample query may be verified, for example, be validated through knowledge base and / or manager, and template generation system 100 'may further comprise components for verification.

In step S90, an operation of generating a template may be performed. For example, the template generation unit 160 may acquire a pattern of ontology components corresponding to a natural language query, and may store an ontology component having a value corresponding to a resource in a sample query as a variable having a type of the ontology component It is possible to generate a template corresponding to the pattern. The generated template may be stored together with the pattern in the template database 900 'and used in query generation according to a natural language query by a query response system (e.g., 300 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 system for generating a template for generating a query from a natural language query to a knowledge base,
A natural language quality preprocessing unit configured to acquire resources each including ontology components corresponding to tokens extracted from the natural language query by transmitting a natural language query to the natural language recognition system;
A graph analyzer configured to generate graph data including the resources based on an ontology schema of the knowledge base composed of knowledge data units and to generate at least one knowledge data unit from the graph data; And
A template generator configured to generate a sample query from the at least one knowledge data unit and generate a template corresponding to the pattern of resources from the sample query. The method according to claim 1,
Wherein the natural language query preprocessor is configured to collect a plurality of natural language queries through a network or to obtain user natural language queries provided to a query response system. The method of claim 2,
Wherein the natural language query preprocessing unit is configured to acquire a natural language query that is not matched with the template in the processing by the query response system among the natural language queries provided to the query response system. The method according to claim 1,
Wherein the graph analyzer comprises: a graph generator configured to generate the graph data by generating a node from a resource including an entity or a class among the ontology components and generating an edge from a resource of the ontology components;
A resource filtering unit configured to filter the resources by selectively removing each of the nodes and edges included in the graph data; And
And a knowledge data unit generation unit configured to generate at least one knowledge data unit from the graph data including the filtered resources. The method of claim 4,
Wherein the resource filtering unit is configured to filter the resources by removing at least one of the node having no edge and the edge having no edge connected to the node. The method of claim 4,
Each of the resources further comprising a weight corresponding to a degree of matching of the token and the ontology component,
Wherein the resource filtering unit is configured to filter the resources by removing interconnected nodes and edges having aggregated weights that are less than or equal to a predefined reference value based on the weights of the nodes and the edges. The method according to claim 1,
Wherein the template generation unit includes a query verification unit configured to verify the sample query based on the knowledge data of the knowledge base. The method of claim 7,
Further comprising an administrator interface,
Wherein the query verifying unit is configured to provide the natural language query and the sample query to the manager through the manager interface and receive information on the suitability and the change of the sample query from the manager through the manager interface . The method according to claim 1,
Wherein the template generation unit is configured to store the generated template in a template storage accessed by a query response system. The method according to claim 1,
Wherein the knowledge data unit is a triple,
Wherein the query is a SPARQL (SPARQL Protocol and RDF Query Language).

Images