KR102673867B1 - Apparatus and method for providing responsive conversation corpus - Google Patents

Abstract

An apparatus and method for providing a corpus of tit-for-tat conversations are disclosed.
A device for providing a matching conversation corpus according to an embodiment of the present invention is a device that provides a matching conversation corpus in conjunction with an example-based conversation system that outputs a response to a user's utterance based on a conversation example corpus. a dialogue example detection unit that detects a dialogue example mapping an utterance and a response corresponding to the utterance in the database of the example-based dialogue system that stores the; a tit-for-tat conversation corpus generating unit that extracts a tit-for-tat response from the responses of the detected conversation examples, and generates a tit-for-tat conversation corpus by mapping the utterances in the detected conversation examples and the extracted tit-for-tat responses; and a matching conversation corpus storage unit that stores the matching conversation corpus.

Description

Apparatus and method for providing responsive conversation corpus}

The present invention relates to an apparatus and method for providing a matching conversation corpus, and more specifically, to a matching conversation corpus that provides a matching conversation corpus in conjunction with an example-based dialogue system that outputs responses to user utterances based on a conversation example corpus. It relates to a provision device and method.

Recently, with the development of voice recognition technology, conversation system technologies that recognize user utterances and provide responses have been introduced.

However, as disclosed in Korean Patent Publication No. 10-2019-0116041, the existing technology provides responses to the user's utterances using a pre-built corpus, so the coverage that can respond to the user's various utterances is limited. There is a problem in that it is limited and the response to the user's utterance is unnatural, which reduces the realism and completeness of the conversation.

In addition, when a large corpus database is separately constructed to increase the response accuracy of the conversation system and expand response coverage, there is a problem that a lot of time and money are required to build the conversation system.

The technical problem that the present invention aims to solve is to enable the dialogue system to selectively output responses for conveying information and counter-argument responses according to the content of the user's speech, thereby simultaneously improving the response coverage and response accuracy of the dialogue system, while simultaneously improving the system's response coverage and response accuracy. The goal is to provide a device and method for providing a conversation corpus that reduces construction costs.

A device for providing a matching conversation corpus according to an embodiment of the present invention is a device that provides a matching conversation corpus in conjunction with an example-based conversation system that outputs a response to a user's utterance based on a conversation example corpus. a dialogue example detection unit that detects a dialogue example mapping an utterance and a response corresponding to the utterance in the database of the example-based dialogue system that stores the; a tit-for-tat conversation corpus generating unit that extracts a tit-for-tat response from the responses of the detected conversation examples, and generates a tit-for-tat conversation corpus by mapping the utterances in the detected conversation examples and the extracted tit-for-tat responses; and a matching conversation corpus storage unit that stores the matching conversation corpus.

In one embodiment, the device creates an input utterance and output response pair that associates an input utterance input to the example-based dialogue system or another dialogue system with an output response output from the corresponding dialogue system in response to the input utterance. It further includes a conversation log storage unit for storing a conversation log database, and the conversation example detection unit is configured to detect an input utterance and output response pair stored in the conversation log database as the conversation example.

In one embodiment, the matching conversation corpus generator separates a response part containing at least one word from the response of the detected conversation example, and the separated response part is included as an independent response in the conversation example corpus. a matching response extractor for extracting the separated response part as the matching response; and a matching response integrator that integrates the grouped matching responses by grouping the matching responses extracted by the matching responses extracting unit and selecting a representative matching response representing the grouped matching responses.

In one embodiment, the matching response extractor is configured to separate a response portion corresponding to the first sentence from the response when the response of the detected conversation example includes a plurality of sentences.

In one embodiment, the matching response extractor is configured to separate, from the response of the detected conversation example, a response portion that includes a plurality of consecutive words but includes words within a predetermined number.

In one embodiment, the response portion including the plurality of consecutive words includes a first word constituting a response to the detected conversation example and at least one word consecutive from the first word.

In one embodiment, the matching conversation corpus generator further includes a matching conversation example generating unit that generates a matching conversation example by mapping the utterance of the detected conversation example and the representative matching response.

In one embodiment, the device uses the tit-for-tat conversation corpus stored by the tit-for-tat conversation corpus storage unit when a response corresponding to the user's utterance input to the example-based dialogue system does not exist in the conversation example corpus. It further includes a matching response output unit that outputs an appropriate matching response corresponding to the user's utterance.

A method of providing a matching conversation corpus according to an embodiment of the present invention is a method of providing a matching conversation corpus in conjunction with an example-based conversation system in which a computer device outputs a response to a user's utterance based on a conversation example corpus, Step (a) of a computer device detecting a conversation example mapping an utterance and a response corresponding to the utterance in the database of the example-based conversation system that stores the conversation example corpus; Step (b) in which the computer device extracts a tit-for-tat response from the response of the detected conversation example, and generates a tit-for-tat conversation corpus by mapping the utterance of the detected conversation example and the extracted tit-for-tat response; and step (c), wherein the computer device stores the corpus of tit-for-tat conversation.

In one embodiment, the method includes, before step (a), the computer device receives an input speech input to the example-based dialogue system or another dialogue system, and an output response output from the dialogue system in response to the input speech. It further includes the step of storing the associated input utterance and output response pair in a conversation log database, wherein step (a) is performed, wherein the computer device detects the input utterance and output response pair stored in the conversation log database as a conversation example. It includes steps to:

In one embodiment, step (b) separates a response part containing at least one word from the response of the detected conversation example, and the separated response part is included as an independent response in the conversation example corpus. step (b1) of extracting the separated response part as the matching response; and a step (b2) of grouping the matching responses extracted by the matching responses extracting unit and selecting a representative matching response representing the grouped matching responses, thereby integrating the grouped matching responses.

In one embodiment, step (b1) includes, when the response of the detected conversation example includes a plurality of sentences, separating a response portion corresponding to the first sentence from the response.

In one embodiment, the step (b1) includes separating a response part containing a plurality of consecutive words from the response of the detected conversation example, but separating a response part containing words within a predetermined number. Includes.

In one embodiment, the response portion including the plurality of consecutive words includes a first word constituting a response to the detected conversation example and at least one word consecutive from the first word.

In one embodiment, the step (b) further includes a step (b3) of generating a tit-for-tat conversation example by mapping the utterance of the detected conversation example and the representative tit-for-tat response.

In one embodiment, the method is such that, when a response corresponding to a user's utterance input to the example-based dialogue system does not exist in the dialogue example corpus, the computer device stores the tit-for-tat dialogue corpus stored in step (c). It further includes step (e) of outputting an appropriate response corresponding to the user's utterance using .

Embodiments according to the present invention can be implemented using a computer program recorded on a recording medium as a computer program that executes the above-described operations or methods through a computer system.

According to the present invention, a device for providing a matching conversation corpus that can be implemented as a computer device or an application automatically generates and provides a matching conversation corpus using the conversation example corpus of an example-based conversation system, allowing the conversation system to follow the content of the user's speech. It allows selective output of responses to convey information and counter-responses, simultaneously improving response coverage and response accuracy of the conversation system, while reducing system construction costs.

In addition, the matching conversation corpus providing device further uses the conversation log recording the conversation history of the conversation system to create a matching conversation corpus, adaptively expanding the response coverage of the conversation system and improving the conversation progress through the conversation system. Reality and completeness can be improved.

Furthermore, those skilled in the art to which the present invention pertains will be able to clearly understand from the following description that various embodiments according to the present invention can solve various technical problems not mentioned above.

1 is a diagram showing an example of a conversation system to which the present invention is applied.
Figure 2 is a block diagram illustrating a device for providing a corpus of tit-for-tat conversation according to an embodiment of the present invention.
Figure 3 is a flowchart showing a method of providing a corpus of tit-for-tat conversation according to an embodiment of the present invention.
Figure 4 is a diagram showing an example of a matching response extraction algorithm applicable to the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to clarify solutions to the technical problems of the present invention. However, in describing the present invention, if the description of related known technology rather obscures the gist of the present invention, the description thereof will be omitted. Additionally, the terms used in this specification are terms defined in consideration of the functions in the present invention, and may vary depending on the intention or custom of the designer, manufacturer, etc. Therefore, definitions of terms described below should be made based on the content throughout this specification.

Figure 1 shows an example of a conversation system 2 to which the present invention is applied.

As shown in FIG. 1, the conversation system 2 to which the present invention is applied may be configured as an example-based conversation system using a corpus of conversation examples.

That is, the conversation system 2 is connected to a smart terminal 10 that recognizes the user's utterance and outputs a voice, such as an artificial intelligence speaker or a smart set-top box, and the smart terminal 10 through a communication network. It may include a conversation service server 20 that outputs a response to the user's utterance. In this case, the conversation service server 20 may include a conversation example corpus database 22 that stores and manages conversation example corpora.

The device for providing a matching conversation corpus according to the present invention automatically generates a matching conversation corpus in conjunction with an example-based conversation system that outputs responses to the user's utterances based on the conversation example corpus as described above, so that the matching conversation corpus is input to the conversation system. It is configured to provide appropriate responses to the user's various utterances.

In the present invention, the ‘joint response’ is not composed of limited words or phrases such as simple chuimsae, interjections, or exclamations, but is composed of sentences with various contents including multiple syllables.

FIG. 2 shows a block diagram of an apparatus 100 for providing a corpus of confrontational dialogue according to an embodiment of the present invention.

As shown in FIG. 2, the device 100 for providing a matching conversation corpus according to an embodiment of the present invention is linked with an example-based conversation system 2 that outputs a response to the user's utterance based on a conversation example corpus. It may be composed of a type of computer device or server, and may include a communication unit 101, an input unit 102, a storage unit 104, an output unit 103, and a control unit 105.

The communication unit 101 receives data transmitted from the example-based dialogue system 2, another dialogue system, or other communication devices through a wired/wireless communication network and transmits it to the control unit 105. It is configured to transmit the control signal or data processed in the example-based conversation system (2), another conversation system, or other communication device. To this end, the communication unit 101 may include a communication modem that performs wired and wireless communication. In this case, the communication unit 101 may include an LTE communication module, a 5G communication module, etc. that perform long-distance wireless communication, a Bluetooth communication module, a Zigbee communication module, a Wi-Fi communication module, etc., that perform short-distance wireless communication. Additionally, the communication unit 101 may include a USB port, a wired LAN port, or other various communication ports to which data transmission cables are connected.

The input unit 102 is configured to receive commands or data from a system operator or administrator. To this end, the input unit 102 may include an input device such as a keyboard, operation button, or touch panel.

The output unit 103 is configured to output data or information processed through the control unit 105 visually or audiovisually. To this end, the output unit 103 may include a visual output device such as a monitor, display panel, or touch screen. Additionally, the output unit 103 may further include a sound generating device such as a speaker.

The storage unit 104 is configured to store and manage data necessary for the operation of the device 100 for providing a conversation corpus. To this end, the storage unit 104 may optionally include various storage media such as ROM, RAM, EEPROM, register, flash memory, CD-ROM, magnetic tape, hard disk, floppy disk, and optical data recording device. Additionally, the storage unit 104 may include a matching conversation corpus database 104a and a conversation log database 104b. In this case, the matching conversation corpus database 104a stores and manages the matching conversation corpus created according to the present invention. The conversation log database 104b stores and manages conversation logs that record utterances input to a conversation system that works with the matching conversation corpus providing device 100 and responses output from the conversation system in response to the utterances.

The control unit 105 is configured to control the overall operation of the device 100 for providing a conversation corpus. To this end, the control unit 105 may optionally include hardware such as a general-purpose processor, an application-specific integrated circuit (ASIC), other chipsets, logic circuits, registers, and memory for executing control logic. Meanwhile, the control unit 105 may be composed of a combination of hardware and software. That is, the control logic of the control unit 105 is composed of a computer program and stored in the internal memory or storage unit 104 of the control unit 105, and the stored computer program may be configured to be executed through the hardware of the control unit 105. .

Meanwhile, the control unit 105 is a functionally divided component and includes a conversation example detection unit 110, a matching conversation corpus generation unit 120, and a matching conversation corpus storage unit 130, and outputs a matching response depending on the embodiment. It may further include a unit 140, a conversation log storage unit 150, etc.

The dialogue example detection unit 110 is configured to detect dialogue examples that map utterances and responses corresponding to the utterances in the database 22 of the example-based dialogue system 2 that stores a corpus of dialogue examples. In general, since example-based dialogue systems use a dialogue example corpus consisting of 100,000 to 10 million utterance (or question)-response pairs (i.e., dialogue examples), the dialogue example detector 110 detects examples. Many and diverse conversation examples can be detected in the database 22 of the base conversation system 2.

The tit-for-tat conversation corpus generation unit 120 extracts a tit-for-tat response from the response of the conversation example detected by the conversation example detection unit 110, and maps the utterance of the detected conversation example to the extracted tit-for-tat response. It is configured to create a corpus. To this end, the matching dialogue corpus generation unit 120 may include a matching dialogue response extraction unit 122, a matching dialogue response integration unit 124, and a matching dialogue example generating unit 126.

In this case, the matching response extractor 122 is configured to separate the response part including at least one word from the response of the detected conversation example and extract the corresponding response part as a matching response.

In the dialogue example corpus of the example-based dialogue system (2), you can obtain dialogue examples consisting of various types of utterances (questions) and various responses to the utterances. In order to extract appropriate matching responses from these various responses, an extraction method using the linguistic characteristics of Korean must be applied. In other words, in Korean conversations, sentences expressing the other person's agreement, sympathy, or additional opinions are usually located at the beginning of the response. Nevertheless, if the entire response is extracted as a matching response, the matching response becomes too long and the coverage of the matching response becomes narrow.

Therefore, in one embodiment, when the response of the conversation example detected by the conversation example detection unit 110 includes a plurality of sentences, the response extractor 122 selects the response part corresponding to the first sentence in the response. It can be configured to separate and extract it as a matching response.

For example, the response in the conversation example is “Merry Christmas, what are you going to do today?” Have a nice day.”, “It sounds really delicious. When a plurality of sentences are divided through punctuation marks, such as “Ramen is always delicious,” the matching response extractor 122 extracts the first sentences from the responses, “Merry Christmas, what are you going to do today?” and “What are you going to do today?” “It must be really delicious.” can be separated and extracted as a tit-for-tat response to that utterance.

In addition, if the response in the conversation example does not include punctuation marks, such as “Buckwheat noodles made from buckwheat are perfect for summer, they are really delicious,” but is divided into multiple sentences through syntactic structure analysis, the correct response extractor 122 Likewise, the first sentence, “Buckwheat noodles made from buckwheat are perfect for summer,” can be separated from the response and extracted as a tit-for-tat response to the utterance.

In another embodiment, the matching response extractor 122 includes a plurality of consecutive words in the response of the conversation example detected by the conversation example detector 110, but includes words within a predetermined number. It can be configured to separate the parts and extract them as a matching response. In this case, the response portion including the plurality of consecutive words may include the first word constituting the response to the detected conversation example and at least one word consecutive from the first word.

For example, the tit-for-tat response is determined to contain less than 5 words, and the response in the conversation example is “Merry Christmas, what are you going to do today?” Have a nice day.”, “It sounds really delicious. In the case of “Ramen is always delicious.”, the tit-for-tat response extractor 122 separates “Merry Christmas” and “It must be really delicious,” which each contain two phrases, from the responses and extracts them as a tit-for-tat response to the corresponding utterance. can do.

On the other hand, if the tit-for-tat response is determined to contain 5 or fewer words, and the response in the conversation example is “Buckwheat noodles made from buckwheat are perfect for summer, they are really delicious,” the tit-for-tat response extractor 122 will say, “In summer, buckwheat noodles are perfect.” “Buckwheat noodles made from buckwheat are perfect” is the first sentence of the response, but since it contains 6 words, it may not be extracted as a tit-for-tat response.

In addition, the tit-for-tat response extractor 122 may be configured to extract the response part as a tit-for-tat response only when the separated response part as described above is included as an independent response in the entire response set of the conversation example corpus. . This is because if the separated response part is included as an independent response in the entire response set of the above conversation example corpus, it can be viewed as a complete response in itself.

The matching response integration unit 124 is configured to group the matching responses extracted by the matching answer extracting unit 122 and integrate the grouped matching responses by selecting a representative matching response representing the grouped matching responses. You can.

Because the responses are in natural language, they often have the same meaning but different forms. Therefore, you can integrate responses with the same meaning through the two methods below and use the integrated response as a matching response. The two methods below have conflicting advantages and disadvantages, so depending on the situation, one of the two methods can be appropriately selected to integrate a tit-for-tat response.

First, the method using clustering is a method that automatically creates clusters of responses using a clustering algorithm and assigns an appropriate matching response to each generated cluster. Each response is expressed as a semantic vector, and by specifying an appropriate tit-for-tat response to the cluster created through the clustering algorithm, all input utterances included in the cluster can be created as a corpus. This method can generate a large amount at once, but there is a possibility that the cluster may contain errors.

The method using high-frequency sorting of matching responses is a method of sorting and outputting matching responses in order of high frequency and integrating matching responses according to the selection of the user or administrator. This method requires more time than the method using clustering, but errors can be reduced because the user or administrator goes through the process of directly selecting and integrating matching responses according to the purpose of the system.

Meanwhile, the matching conversation example generation unit 126 is configured to generate a matching conversation example by mapping the utterance of the conversation example detected by the conversation example detection unit 110 and the matching response extracted by the matching response extracting unit 122. do. In this case, the matching conversation example generating unit 126 may be configured to generate matching conversation examples by mapping the utterance of the detected conversation example and the representative matching response selected by the matching conversation example integration unit 124. In this way, by constructing a tit-for-tat conversation corpus with examples of tit-for-tat dialogues created by mutually mapping various types of utterances and a small amount of tit-for-tat responses selected through a tit-for-tat response integration process, the data size of the tit-for-tat conversation corpus is reduced while The coverage of the response can be expanded.

The matching conversation corpus storage unit 130 is configured to store the matching conversation corpus including the matching conversation examples created as described above in the database 104a.

The matching response output unit 140 is configured to output a matching response in response to the user's utterance input to the conversation system that works with the matching conversation corpus providing device 100. For example, if the response corresponding to the user's utterance input to the example-based dialogue system 2 does not exist in the dialogue example corpus of the corresponding dialogue system, the matching response output unit 140 uses the matching dialogue corpus to It may be configured to output the most appropriate tit-for-tat response corresponding to the user's utterance among a small amount of tit-for-tat responses included in the conversation corpus.

The conversation log storage unit 150 stores input utterances and output responses that associate input utterances input to the example-based dialog system 2 or other dialog systems with output responses output from the corresponding dialog system in response to the input utterances. It is configured to store the pair in the conversation log database 104b. In this case, the conversation example detection unit 110 may be configured to detect the input utterance and output response pair stored in the conversation log database 104b as a conversation example for extracting a matching response. As a result, the matching conversation corpus generator 120 can extract more diverse matching responses using the input utterance and output response pairs detected in the conversation log database 104b.

In this way, when an example-based conversation system using a conversation example corpus has already been built, the matching conversation corpus 100 according to the present invention can be used to cost-effectively generate and provide a matching conversation corpus.

The components of the control unit 105 described above may be integrated with other components or may be subdivided into sub-components for step-by-step execution of control logic. Even if the components of the control unit 105 are integrated or segmented, if the same function is recognized, the integrated or segmented components are also included in the scope of the present invention.

In addition, in Figure 2, the device 100 for providing a matching conversation corpus according to the present invention is described as being composed of a separate device from the example-based conversation system 2, but according to the embodiment, the device 100 for providing a matching conversation corpus is provided as an example. It may also be integrated with the base conversation system (2).

Figure 3 is a flowchart showing a method of providing a corpus of tit-for-tat conversation according to an embodiment of the present invention. With reference to FIG. 3 , detailed operations of the device 100 for providing a conversation corpus will be described in chronological order.

As shown in FIG. 3, first, the dialogue example detection unit 110 of the device 100 detects an utterance and a corresponding utterance in the database 22 of the example-based dialogue system 2 that stores a corpus of dialogue examples. A conversation example mapping a response is detected (S310).

In this case, the conversation log storage unit 150 of the device 100 stores input speech previously input to the example-based dialogue system 2 or another dialogue system, and output from the dialogue system in response to the input speech. The input utterance and output response pairs associated with the output response are stored in the conversation log database 104b, and the conversation example detector 110 extracts a matching response from the input utterance and output response pair stored in the conversation log database 104b. It can be detected as a conversation example for .

Next, the tit-for-tat conversation corpus generator 120 of the device 100 extracts a tit-for-tat response from the response to the conversation example detected by the conversation example detection unit 110, and extracts and utterances the detected conversation example. A corpus of tit-for-tat conversations is created by mapping the tit-for-tat responses (S320 to S340).

In this case, the matching response extractor 122 of the matching conversation corpus generation unit 120 separates the response part containing at least one word from the response of the conversation example detected by the conversation example detection unit 110 and responds to the corresponding response. The part can be extracted as a matching response (S320).

In one embodiment, when the response of the conversation example detected by the conversation example detection unit 110 includes a plurality of sentences, the matching response extractor 122 separates the response part corresponding to the first sentence from the response. Thus, it can be extracted as a matching response.

In another embodiment, the matching response extractor 122 includes a plurality of consecutive words in the response of the conversation example detected by the conversation example detector 110, but includes words within a predetermined number. It can be configured to separate the parts and extract them as a matching response. In this case, the response portion including the plurality of consecutive words may include the first word constituting the response to the detected conversation example and at least one word consecutive from the first word.

Figure 4 shows an example of a matching response extraction algorithm applicable to the present invention. In Figure 4, S is the entire response set of the conversation example corpus, L is the maximum number of words included in the correct response, and Answer[0:i] is the response part including up to the i-th word in the response.

As shown in FIG. 4, the matching response extractor 122 can extract a matching response from the corresponding response using a subset algorithm. That is, the matching response extractor 122 separates the response part including up to the i-th word from the corresponding response, and the separated response part is included as an independent response in the entire response set (S) of the conversation example corpus. , the corresponding response part can be extracted as a matching response.

Referring again to FIG. 3, the matching response integration unit 124 of the matching conversation corpus generation unit 120 groups the matching responses extracted by the matching matching response extracting unit 122, and generates a grouping of matching responses representing the grouped matching responses. The grouped matching responses can be integrated by selecting a representative matching response (S330).

As mentioned above, the matching response integrator 124 may integrate the matching responses using a method using clustering or a method using a high-frequency sorting of the matching responses. That is, the matching response integrator 124 automatically creates clusters of responses using a clustering algorithm and selects an appropriate matching response for each generated cluster to integrate the matching responses, or ranks the matching responses in high frequency order. You can sort and output them and integrate the matching responses as selected by the user or administrator.

Next, the matching conversation example generation unit 126 of the matching conversation corpus generation unit 120 selects the utterance of the conversation example detected by the conversation example detection unit 110 and the matching conversation example extracted by the matching response extracting unit 122. A tit-for-tat conversation example can be created by mapping the response (S340). In this case, the matching conversation example generating unit 126 may generate matching conversation examples by mapping the utterance of the detected conversation example and the representative matching response selected by the matching conversation example integration unit 124.

Next, the matching conversation corpus storage unit 130 of the device 100 stores the matching conversation corpus including the matching conversation examples generated as described above in the database 104a (S350).

Then, the tit-for-tat response output unit 140 of the device 100 may output a tit-for-tat response in response to the user's utterance input into the conversation system that works with the chat corpus providing device 100 (S360) ). For example, if the response corresponding to the user's utterance input to the example-based dialogue system 2 does not exist in the dialogue example corpus of the corresponding dialogue system, the matching response output unit 140 uses the matching dialogue corpus to The most appropriate tit-for-tat response corresponding to the user's utterance can be output among a small amount of tit-for-tat responses included in the tit-for-tat conversation corpus.

Meanwhile, embodiments according to the present invention may be implemented with a computer system and a computer program that runs the computer system. When embodiments of the present invention are implemented as a computer program, the components of the present invention are program segments that execute the corresponding operation or task through the computer system. These computer programs or program segments may be stored in various computer-readable recording media. Computer-readable recording media include all types of media that record data that can be read by a computer system. For example, computer-readable recording media may include ROM, RAM, EEPROM, registers, flash memory, CD-ROM, magnetic tape, hard disk, floppy disk, or optical data recording device. Additionally, these recording media can be distributed across computer systems connected to various networks to store or execute program codes in a distributed manner.

As described above, according to the present invention, a device for providing a matching conversation corpus that can be implemented as a computer device or an application automatically generates and provides a matching conversation corpus using a conversation example corpus of an example-based conversation system, It allows the dialogue system to selectively output responses to convey information and counter-argument responses depending on the content of the user's speech, and simultaneously improves the response coverage and response accuracy of the dialogue system, while reducing system construction costs.

In addition, the matching conversation corpus providing device further uses the conversation log recording the conversation history of the conversation system to create a matching conversation corpus, adaptively expanding the response coverage of the conversation system and improving the conversation progress through the conversation system. Reality and completeness can be improved.

Furthermore, it goes without saying that the embodiments according to the present invention can solve various technical problems other than those mentioned in this specification in the relevant technical field as well as in the related technical field.

So far, the present invention has been described with reference to specific embodiments. However, those skilled in the art will clearly understand that various modified embodiments can be implemented within the technical scope of the present invention. Therefore, the previously disclosed embodiments should be considered from an explanatory perspective rather than a limiting perspective. In other words, the scope of the true technical idea of the present invention is shown in the claims, and all differences within the scope of equivalents should be construed as being included in the present invention.

110: Conversation example detection unit 120: Conversation conversation corpus generation unit
122: Matching response extraction unit 124: Matching response integration unit
126: Matching conversation example generation unit 130: Matching conversation corpus storage unit
140: response output unit 150: conversation log storage unit

Claims (17)

A tit-for-tat conversation corpus providing device that provides a tit-for-tat conversation corpus in conjunction with an example-based dialogue system that outputs the system's response to the user's utterance based on a conversation example corpus,
a dialogue example detection unit that detects a dialogue example mapping a user's utterance and a response of a system corresponding to the utterance from the database of the example-based dialogue system that stores the dialogue example corpus;
a tit-for-tat conversation corpus generating unit that extracts a tit-for-tat response from the system response of the detected conversation example, and generates a tit-for-tat conversation corpus by mapping the user's utterance of the detected conversation example and the extracted tit-for-tat response;
a matching conversation corpus storage unit that stores the matching conversation corpus; and
If the system response corresponding to the user's utterance input to the example-based dialogue system does not exist in the dialogue example corpus, it is input to the example-based dialogue system using the tit-for-tat dialogue corpus stored by the tit-for-tat dialogue corpus storage unit. A matching dialogue corpus providing device comprising: a matching response output unit that outputs a matching response corresponding to the user's utterance. According to paragraph 1,
The device stores an input utterance and an output response pair that associate an input utterance input to the example-based dialogue system or another dialogue system and an output response output from the dialogue system in response to the input utterance in a dialogue log database. Further comprising a conversation log storage unit,
The conversation example detection unit is configured to detect input utterance and output response pairs stored in the conversation log database as conversation examples. According to paragraph 1,
The tit-for-tat conversation corpus generation unit,
A response part containing at least one word is separated from the system response of the detected conversation example, and if the separated response part is included as an independent response in the dialogue example corpus, the separated response part is converted into the matching response. A matching response extraction unit for extracting; and
Providing a tit-for-tat conversation corpus, comprising a tit-for-tat response integration unit that groups the tit-for-tat responses extracted by the tit-for-tat response extractor and integrates the grouped tit-for-tat responses by selecting a representative tit-for-tat response representing the grouped tit-for-tat responses. Device. According to clause 3,
The tit-for-tat response extractor is configured to separate a response part corresponding to the first sentence from the response when the response of the system of the detected conversation example includes a plurality of sentences. According to paragraph 3,
The tit-for-tat response extractor is configured to separate, from the system response of the detected conversation example, a response portion that includes a plurality of consecutive words but includes words within a predetermined number. According to clause 5,
The response part including the plurality of consecutive words includes a first word constituting the system's response to the detected conversation example and at least one word consecutive from the first word. Device. According to paragraph 3,
The tit-for-tat conversation corpus generating unit further comprises a tit-for-tat conversation example generating unit that generates a tit-for-tat conversation example by mapping the user's utterance of the detected conversation example and the representative tit-for-tat response. delete A method of providing a matching conversation corpus in which a computer device provides a matching conversation corpus in conjunction with an example-based conversation system that outputs the system's response to the user's utterance based on a conversation example corpus, comprising:
Step (a) of detecting a conversation example mapping a user's utterance and a response of a system corresponding to the utterance in the database of the example-based conversation system where the computer device stores the conversation example corpus;
Step (b) in which the computer device extracts a tit-for-tat response from the system response of the detected conversation example, and generates a tit-for-tat conversation corpus by mapping the user's utterance of the detected conversation example and the extracted tit-for-tat response;
Step (c) of the computer device storing the corpus of tit-for-tat conversations; and
If the system response corresponding to the user's utterance input to the example-based dialogue system does not exist in the dialogue example corpus, the computer device uses the tit-for-tat dialogue corpus to determine the user's utterance input to the example-based dialogue system. A method of providing a matching conversation corpus comprising: outputting a matching response corresponding to. According to clause 9,
The method includes an input utterance in which the computer device associates an input utterance input to the example-based dialogue system or another dialogue system with an output response output from the corresponding dialogue system in response to the input utterance before step (a), and Further comprising the step of storing the output response pair in a conversation log database,
The step (a) includes detecting, by the computer device, an input utterance and output response pair stored in the conversation log database as a conversation example. According to clause 9,
In step (b),
A response part containing at least one word is separated from the system response of the detected conversation example, and if the separated response part is included as an independent response in the dialogue example corpus, the separated response part is converted into the matching response. Extracting (b1) step; and
A corpus of tit-for-tat conversations comprising a step (b2) of grouping the tit-for-tat responses extracted by the tit-for-tat response extractor and integrating the grouped tit-for-tat responses by selecting a representative tit-for-tat response representing the grouped tit-for-tat responses. How to provide. According to clause 11,
The step (b1) includes, when the response of the system of the detected conversation example includes a plurality of sentences, separating the response part corresponding to the first sentence from the response. A corpus of tit-for-tat conversation. How to provide. According to clause 11,
The step (b1) includes the step of separating a response part containing a plurality of consecutive words from the system response of the detected conversation example, but separating a response part containing words within a predetermined number. A method of providing a corpus of characteristic tit-for-tat conversations. According to clause 13,
The response part including the plurality of consecutive words includes a first word constituting the system's response to the detected conversation example and at least one word consecutive from the first word. method. According to clause 11,
The step (b) further includes a step (b3) of generating a tit-for-tat conversation example by mapping the user's utterance of the detected conversation example and the representative tit-for-tat response. delete A computer program that executes the method according to any one of claims 9 to 15 through a computer, and is recorded on a computer-readable recording medium.

Images