KR20220026890A - Method and apparatus for generating ontology based on dialog - Google Patents

Abstract

Provided is a method for generating the ontology, which includes the steps of: receiving input of the conversation; recognizing a conversation situation including at least one of a time, a place, and a case in which the conversation occurred; determining the higher level knowledge corresponding to the conversation situation, and constructing the higher level knowledge ontology using the higher level knowledge; and searching for the lower level knowledge ontology corresponding to the conversation situation, and linking the lower level knowledge ontology to the higher level knowledge ontology.

Description

Method and device for generating dialogue-based ontology

The present invention relates to a technology for generating an ontology based on a conversation.

The service-related ontology used in the conventional recommendation system, etc. is mainly generated before the service by using an ontology authoring tool by an expert with domain knowledge about the service. By using the generated ontology, knowledge or information for creating a conversation was provided. Afterwards, the generated ontology was modified based on the information collected during the service process.

Recently, a lot of conversational data between artificial intelligence machines and users is generated through interactive service systems such as chatbots. Since a lot of domain knowledge is inherent in such a conversation, if the ontology is supplemented by extracting knowledge from the generated conversation in reverse, a service can be effectively provided.

However, in the service ontology created by experts, it is difficult to reflect the knowledge collected from the dialogue provided by the dialogue system from moment to moment, and it is difficult to provide effective services for errors in knowledge or changing knowledge.

Accordingly, there is a need for a technology capable of generating or updating an ontology by extracting knowledge information from conversations collected in a conversation system in service.

An object to be solved is to provide a method and apparatus for generating an ontology by extracting service knowledge from a conversation between a system and a user in a dialog system and updating the stored ontology.

In addition, the task to be solved is to provide a method and apparatus for identifying a conversation situation and generating a response by inferring service knowledge corresponding thereto.

A method for generating an ontology according to an embodiment is disclosed. The method includes the steps of receiving a conversation input, recognizing a conversation situation including at least one of a time, place, and case in which the conversation occurred, determining a higher level knowledge corresponding to the conversation situation, and using the higher level knowledge Constructing a higher-level knowledge ontology, and searching for a lower-level knowledge ontology corresponding to the dialogue situation, and linking the lower-level knowledge ontology to the higher-level knowledge ontology.

According to the present invention, unlike the existing dialog system in which a service ontology is created by an expert, it is possible to provide a faster and more accurate dialog service by using data obtained from the dialog system to directly create a service ontology.

In addition, according to the present invention, since knowledge inherent in a conversation is quickly converted into an ontology, an improved service can be provided to the user as the conversation progresses.

In addition, according to the present invention, by using a user profile, knowledge extracted from conversations with each user can be created as an ontology to provide a personalized service.

1 is a block diagram of an ontology generating apparatus according to an embodiment.
2 is an exemplary diagram of an ontology for each service according to an embodiment.
3 is an explanatory diagram of a method for generating service knowledge according to an embodiment.
4 is a flowchart of an ontology update method according to an embodiment.
5 is a flowchart of a method for generating a response according to an embodiment.
6 is a diagram illustrating a computer system according to an embodiment.

Hereinafter, with reference to the accompanying drawings, the embodiments of the present invention will be described in detail so that those of ordinary skill in the art to which the present invention pertains can easily implement them. However, the present invention may be embodied in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Ontology means that the relationship between entities is defined and expressed according to certain rules so that the computer can understand the semantics of all entities.

Higher-level knowledge refers to general concepts such as time, space, material, object, and event that are not related to specific problems or specific areas of interest, and upper-level knowledge ontology refers to an ontology that describes higher-level knowledge.

The sub-knowledge refers to a concept related to a specific task, topic, or service, and the sub-knowledge ontology refers to an ontology expressed about the sub-knowledge.

1 is a block diagram of an ontology generating apparatus according to an embodiment, FIG. 2 is an exemplary diagram of an ontology for each service according to an embodiment, and FIG. 3 is an explanatory diagram of a method for generating service knowledge according to an embodiment .

Referring to FIG. 1 , the ontology generating apparatus 100 receives a text or voice dialogue input, extracts a dialogue situation occurring in the process of providing a response thereto, and generates higher-level knowledge as an ontology from the extracted situation, , the lower knowledge corresponding to the generated upper knowledge ontology is connected and stored.

In addition, the ontology generating apparatus 100 extracts the context of the input conversation, searches for knowledge appropriate to the context and ontology for each service, and generates and provides a response based on this.

The ontology generating apparatus 100 includes an input/output unit 110 , a preprocessor 120 , a higher knowledge generating unit 130 , a lower knowledge generating unit 140 , an ontology storage unit 150 , a response generating unit 160 , and a knowledge It includes an inference unit 170 .

For explanation, the input/output unit 110 , the preprocessor 120 , the upper knowledge generation unit 130 , the lower knowledge generation unit 140 , the ontology storage unit 150 , the response generation unit 160 , and the knowledge inference unit 170 . ), but these are computing devices operated by at least one processor. Here, the input/output unit 110 , the preprocessor 120 , the upper knowledge generation unit 130 , the lower knowledge generation unit 140 , the ontology storage unit 150 , the response generation unit 160 , and the knowledge inference unit 170 ) may be implemented in one computing device or distributed in separate computing devices. When distributed and implemented in a separate computing device, the input/output unit 110 , the preprocessor 120 , the upper knowledge generation unit 130 , the lower knowledge generation unit 140 , the ontology storage unit 150 , and the response generation unit 160 . ) and the knowledge reasoning unit 170 may communicate with each other through a communication interface. The computing device may be any device capable of executing a software program written to carry out the present invention, and may be, for example, a server, a laptop computer, or the like.

The input/output unit 110 receives a text or voice dialogue, provides it to the preprocessor 120 and the response generation unit 160 , and outputs the response received from the response generation unit 160 . The form of the response may vary depending on the type of service to which the ontology generating apparatus 100 is applied. For example, in the case of a conversation system, the response may be a text or voice conversation, and in the case of a recommendation system, it may be in the form of an image or a URL.

The pre-processing unit 120 generates learning data by processing the conversations collected by the input/output unit 110 , and transmits it to the upper knowledge generation unit 130 .

The preprocessor 120 may generate learning data as one conversation bundle by identifying the start and end of a conversation exchanged between the user and the ontology generating apparatus 100 . The unit of the bundle may be one episode or a conversation between the user and the ontology generating apparatus 100 once each.

As an example, when the ontology generating device 100 is applied to a task-oriented dialog system, the dialog period is standardized and constant, so it is possible to generate the dialog until the task is completed as learning data. there is.

As another example, when applied to a non-task-oriented conversation system such as a chatbot, learning data may be generated according to the topic of the conversation. A method of generating the training data is not limited to any one.

The upper knowledge generating unit 130 receives the learning data, determines the situation of a conversation included in the learning data, and generates upper knowledge appropriate to the situation to configure the upper knowledge ontology.

Then, the lower knowledge ontology corresponding to the determined situation is searched for in the ontology storage unit 150, and the lower knowledge ontology is connected to the generated upper knowledge ontology and stored.

On the other hand, the situation of the conversation is determined by the situation factors, and the situation factors include, for example, a time (Time), a place (Place), and an occasion (Occasion) in which the conversation takes place. That is, the upper knowledge generating unit 130 identifies when, where, and in what cases the conversations constituting the learning data took place. According to the learning data, a combination of individual context elements may be defined as a new context element.

Meanwhile, the method for grasping the situation is not limited to any one, and methods such as clustering and pattern matching may be used.

As a conversational situation, possible things are not a fixed set, and may be added or changed according to the learning data. For example, a new situation may be defined by a combination of situation elements extracted from dialogue data.

The higher-order knowledge generator 130 may search for higher-order knowledge corresponding to the determined situation in the ontology storage 150 .

The upper knowledge generating unit 130 determines the situation of the conversation included in the learning data, and calls the lower level knowledge corresponding to this situation to the ontology storage unit 150 to obtain it. Thereafter, the upper knowledge generated in the context of the conversation is connected with the acquired lower knowledge, and again stored in the ontology storage 150 . In this case, the lower knowledge ontology may be written in OWL, and the upper knowledge ontology may be written in SWRL. For example, it may have the form shown in Table 1.

[isWedding: fo:Dress(?a)
^ fo:hasBody_Length(?a, fr:NormalLength)
^ fo:hasGender(?a, fr:Elegant)
^ fo:hasWeight(?a, fr:Weight_Light)
^ fo:hasWristEdge(?a, fr:Lace) -> fo:Wedding(?a)]

Referring to Table 1, it is an example of upper knowledge matching lower knowledge for a situation corresponding to a wedding ceremony. For example, Table 1 shows the rules for designating elements such as Dress and Body_Length of the lower knowledge that describe the upper knowledge of isWedding() (guest fashion).

On the other hand, a specific method for the higher-level knowledge generating unit 130 to grasp the situation and determine the higher-level knowledge is as follows.

When the dialog included in the learning data is in the form of a conversation between the user and the system alternately, the dialog of each conversation subject can be regarded as one information source. In this case, each dialogue may be converted into an S-P-O (Subject-Predicate-Object) form. By tagging the situation in each dialogue, it is possible to make it knowledgeable, and based on this, higher knowledge can be derived.

As another example, the upper knowledge generating unit 130 extracts dialogue acts from a conversation. Speech acts mean the intention of a conversational subject, and a method of analyzing a dialogue act is not limited to any one. Among various types of dialogue acts, since information contains the most knowledge, a dialogue having the dialogue act of information provision can be transformed into an S-P-O form. In addition, higher-level knowledge may be constructed using a dialogue having dialogue acts such as information requests (Ask_ref, Ask_if). On the other hand, dialogue acts such as response and response can be utilized for the user's personalized information.

When the knowledge generated by the upper knowledge generator 130 collides with the knowledge stored in the ontology storage 150 , an adjustment may be made for this. For example, when the generated knowledge collides with the stored knowledge, the upper knowledge generating unit 130 may delete the conflicting knowledge or select and store only the high frequency of appearance among them.

In Table 1, when each knowledge element is connected with ^ (and), if opposite sub-knowledge is selected, the two knowledge can be removed, or the method with a high frequency of appearance can be selected by storing the frequency of appearance of each knowledge element. there is.

The lower knowledge generating unit 140 receives an ontology for each service and stores it in the ontology storage unit 150 . The input ontology may be modified to fit the form managed by the ontology storage unit 150 .

The input ontology may be composed of knowledge used in a specific domain or a specific service. For example, a fashion coordination service ontology of the form shown in FIG. 2 may be input. Specifically, the ontology of FIG. 2 is generated by creating an ontology of individual properties of a fashion item.

The sub-knowledge generating unit 140 may directly generate an ontology by receiving information rather than a completed ontology. In this case, the sub-knowledge generator 140 may include an authoring tool for creating an ontology.

The ontology storage unit 150 stores the ontology provided through the upper knowledge generating unit 130 or the lower knowledge generating unit 140 . The higher knowledge ontology or lower knowledge ontology of a specific service is provided in response to a request from the higher knowledge generating unit 130 , the lower knowledge generating unit 140 , or the knowledge inference unit 170 .

When the user's conversation is input from the input/output unit 110 , the response generating unit 160 analyzes the dialogue act and situation of the input dialogue and provides it to the knowledge inference unit 170 , and the knowledge inference unit 170 matches the situation It receives knowledge for each service and generates a response based on the knowledge. The response may be an image such as a text or voice-type sentence or a recommended item.

Meanwhile, the response generation unit 160 may analyze the current conversation situation as situation elements such as time, place, and case and provide it to the knowledge inference unit 170 , and a method or situation in which the response generation unit 160 analyzes the dialogue act A method of analyzing may be similar to an operation method of the upper knowledge generating unit 130 .

The knowledge inference unit 170 receives the dialogue act and situation of the conversation from the response generating unit 160 , and searches the ontology storage unit 150 for higher-level knowledge appropriate to the situation. Service knowledge corresponding to the conversation may be inferred from the searched higher knowledge and provided to the response generating unit 160 .

In this case, the knowledge inference unit 170 may search for knowledge for each context element of the current conversation and extract service knowledge by combining them.

Referring to FIG. 3 , the knowledge inference unit 170 searches for time-based knowledge, place-based knowledge, and case-based knowledge in the ontology storage unit 150 . In addition, it is possible to search for a plurality of context-based knowledge by considering time and place in a complex way. The knowledge inference unit 170 may construct service knowledge of the current conversation as the intersection or union of the found knowledge.

On the other hand, in general, for individual knowledge such as time-based knowledge, place-based knowledge, and case-based knowledge, higher-level knowledge can be generated based on general common sense, and situation-based knowledge can generate higher-level knowledge by considering special cases in the context of a conversation. there is.

4 is a flowchart of an ontology update method according to an embodiment.

Referring to FIG. 4 , the ontology generating apparatus 100 generates learning data by grouping the received conversations into one unit ( S110 ). Specifically, the preprocessor 120 may determine a unit by identifying the start and end of the conversation.

The ontology generating apparatus 100 extracts a time, a place, and a case from each conversation constituting the learning data to understand the conversation situation ( S120 ). Methods such as clustering, pattern matching, etc. may be used to understand the conversation situation.

The ontology generating apparatus 100 generates higher-order knowledge appropriate to the situation as a higher-level knowledge ontology (S130). For example, the upper knowledge ontology can be written in SWRL.

The ontology generating apparatus 100 searches for a lower-level knowledge ontology appropriate to the situation, and stores it in connection with the generated upper-level knowledge ontology ( S140 ). The sub-knowledge is input through the sub-knowledge generating unit 140 , and may be stored in the ontology storage unit 150 .

5 is a flowchart of a method for generating a response according to an embodiment.

Referring to FIG. 5 , the ontology generating apparatus 100 analyzes the dialogue act of the received dialogue, extracts the time, place, and case of the dialogue to determine the situation of the dialogue ( S210 ).

The ontology generating apparatus 100 searches for an upper knowledge ontology corresponding to a dialogue act and situation of a conversation (S220).

The ontology generating apparatus 100 extracts service knowledge corresponding to the received conversation by using the upper knowledge ontology (S230).

6 is a diagram illustrating a computer system 200 according to an embodiment of the present invention.

Order prediction apparatus according to an embodiment of the present invention may be implemented as a computer system 200 as shown in FIG. In addition, each component of the order prediction device may be implemented as a computer system 200 as shown in FIG. 6 .

The computer system 200 includes at least one of a processor 210 , a memory 220 , a user interface input device 240 , a user interface output device 250 , and a storage device 230 communicating via a bus 260 . may include

The processor 210 may be a central processing unit (CPU) or a semiconductor device that executes instructions stored in the memory 220 or the storage device 230 . The processor 210 may be configured to implement the functions and methods described with reference to FIGS. 1 to 5 .

The memory 220 and the storage device 230 may include various types of volatile or nonvolatile storage media. For example, the memory 220 may include a read only memory (ROM) 221 and a random access memory (RAM) 222 . In an embodiment of the present invention, the memory 220 may be located inside or outside the processor 210 , and the memory 220 may be connected to the processor 210 through various known means.

According to the present invention, unlike the existing dialog system in which a service ontology is created by an expert, it is possible to provide a faster and more accurate dialog service by using data obtained from the dialog system to directly create a service ontology.

In addition, according to the present invention, since knowledge inherent in a conversation is quickly converted into an ontology, an improved service can be provided to the user as the conversation progresses.

In addition, according to the present invention, by using a user profile, knowledge extracted from conversations with each user can be created as an ontology to provide a personalized service.

The embodiment of the present invention described above is not implemented only through the apparatus and method, and may be implemented through a program for realizing a function corresponding to the configuration of the embodiment of the present invention or a recording medium in which the program is recorded.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto. is within the scope of the right.

Claims (1)

receiving input from the conversation;
Recognizing a conversation situation including at least one of a time, a place, and a case in which the conversation occurred;
determining a higher level knowledge corresponding to the dialogue situation, and constructing a higher level knowledge ontology using the higher level knowledge; and
retrieving a lower-level knowledge ontology corresponding to the dialogue situation, and linking the lower-level knowledge ontology to the upper-level knowledge ontology.
Including, an ontology creation method.

Images