KR20230076012A - Method and system for creating persona conversation data using hyperscale language model - Google Patents

Abstract

The present disclosure relates to a conversation database construction method using a language model, which is performed by at least one processor. The conversation database construction method using a language model includes the steps of: receiving a plurality of predefined conversation purpose-conversation type pairs; generating a plurality of conversation situations related to each conversation purpose using a language model; and constructing an initial conversation database by generating a plurality of seed conversation sessions corresponding to a plurality of conversation purpose-conversation type-conversation situation triple pairs using the language model. Each seed conversation session includes at least one user utterance and at least one character utterance.

Description

Method and system for creating persona conversation data using super-large language model

The present disclosure relates to a method and system for generating persona conversation data using a super-large language model, and more specifically, to a method and system for generating persona conversation data capable of finely controlling the conversation flow of a character chatbot.

If a user wants to be provided with information related to a specific service, the user can obtain the desired information by exchanging questions and answers through a professional counselor. However, due to the recent rise in labor costs and the need to reduce waiting time for counseling, artificial intelligence chatbot services that provide answers to user's questions in place of the role of professional counselors are increasing. Meanwhile, in order to provide the same experience as having a conversation with a person rather than a machine, efforts are being made to develop a Persona chatbot service.

In order to solve the above problem, various embodiments of the present disclosure provide a method for constructing a conversation database, a computer program stored in a recording medium, and an apparatus (system).

The present disclosure may be implemented in a variety of ways, including a method, apparatus (system) or computer program stored on a readable storage medium.

According to an embodiment of the present disclosure, a method for constructing a conversation database using a language model, performed by at least one processor, is provided. A method for constructing a conversation database includes the steps of receiving a plurality of predefined conversation purpose-conversation type pairs, generating a plurality of conversation situations associated with each conversation purpose by using a language model, and using a language model to create a plurality of conversation purpose-conversation type pairs. and constructing an initial dialogue database by creating a plurality of seed dialogue sessions corresponding to the dialogue type-dialogue situation triplets, each seed dialogue session including at least one user utterance and at least one character utterance.

According to one embodiment of the present disclosure, a computer program stored in a computer-readable recording medium is provided to execute a conversation database construction method on a computer.

According to one embodiment of the present disclosure, a dialogue database building system is provided. A dialogue database construction system includes a communication module, a memory, and at least one processor connected to the memory and configured to execute at least one computer-readable program included in the memory, wherein the at least one program includes a plurality of predefined conversations. A plurality of seeds corresponding to the plurality of conversation purpose-conversation type-conversation situation triplets are received using the purpose-conversation type pair, and a plurality of conversation situations associated with each conversation purpose are generated using the language model. and instructions for creating a chat session to build an initial dialogue database, each seed chat session including at least one user utterance and at least one character utterance.

According to various embodiments of the present disclosure, by generating persona conversation data using a conversation purpose and a conversation type corresponding to the conversation purpose, the flow of conversation can be controlled in detail while maintaining the linguistic characteristics of the character chatbot. You can provide a character chatbot that reveals the character's personality well.

According to various embodiments of the present disclosure, by generating persona conversation data using a conversation situation, a natural conversation can be continued by providing appropriate speech of a character chatbot in response to user speech in various situations.

According to various embodiments of the present disclosure, a large amount of persona conversation data for generating a character chatbot can be easily and quickly generated by generating a plurality of augmented conversation sessions based on one seed conversation session using a super-large language model. there is.

According to various embodiments of the present disclosure, a high-quality character conversation database can be constructed without excessive human efforts by automatically inspecting generated persona conversation data using a character persona detector.

The effects of the present disclosure are not limited to the effects mentioned above, and other effects not mentioned are clear to those skilled in the art (referred to as "ordinary technicians") from the description of the claims. will be understandable.

1 is a diagram illustrating an example of constructing a final dialogue database including character dialogues using a language model according to an embodiment of the present disclosure.
2 is a schematic diagram showing a configuration in which an information processing system is communicatively connected with a plurality of user terminals in order to provide a character chatbot service according to an embodiment of the present disclosure.
3 is a block diagram illustrating an internal configuration and a very large language model of a user terminal and an information processing system according to an embodiment of the present disclosure.
4 is a diagram illustrating an example of generating a plurality of conversation purpose-conversation type pairs, according to one embodiment.
5 is a diagram illustrating an example of generating a schematic conversation situation, according to an exemplary embodiment.
6 is a diagram illustrating an example of generating a specific conversation situation according to an exemplary embodiment.
7 is a diagram illustrating an example of a prompt for creating a seed conversation session, according to one embodiment.
8 is a diagram illustrating an example of generating an augmented conversation session corresponding to a seed conversation session using a super-large language model according to an embodiment of the present disclosure.
9 is a diagram illustrating an example of a predefined character personality map according to an embodiment of the present disclosure.
10 is a diagram illustrating an example of generating a final conversation database using a character persona detector according to an embodiment of the present disclosure.
11 is a diagram illustrating an example of filtering a conversation session using a personality filter according to an embodiment of the present disclosure.
12 is a diagram illustrating an example of filtering an augmented conversation session using a world view filter according to an embodiment of the present disclosure.
13 is a diagram illustrating an example of converting a character utterance that does not match a character dialogue into a character dialogue using a dialogue filter and a dialogue converter according to an embodiment of the present disclosure.
14 is a flowchart illustrating a method of constructing a dialogue database using a language model according to an embodiment of the present disclosure.

Hereinafter, specific details for the implementation of the present disclosure will be described in detail with reference to the accompanying drawings. However, in the following description, if there is a risk of unnecessarily obscuring the gist of the present disclosure, detailed descriptions of well-known functions or configurations will be omitted.

In the accompanying drawings, identical or corresponding elements are given the same reference numerals. In addition, in the description of the following embodiments, overlapping descriptions of the same or corresponding components may be omitted. However, omission of a description of a component does not intend that such a component is not included in an embodiment.

Advantages and features of the disclosed embodiments, and methods of achieving them, will become apparent with reference to the following embodiments in conjunction with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms, but only the present embodiments make the present disclosure complete, and the present disclosure does not extend the scope of the invention to those skilled in the art. It is provided only for complete information.

Terms used in this specification will be briefly described, and the disclosed embodiments will be described in detail. The terms used in this specification have been selected from general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but they may vary according to the intention of a person skilled in the related field, a precedent, or the emergence of new technology. In addition, in a specific case, there is also a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the invention. Therefore, the terms used in the present disclosure should be defined based on the meaning of the terms and the general content of the present disclosure, not simply the names of the terms.

Expressions in the singular number in this specification include plural expressions unless the context clearly dictates that they are singular. Also, plural expressions include singular expressions unless the context clearly specifies that they are plural. When it is said that a certain part includes a certain component in the entire specification, this means that it may further include other components without excluding other components unless otherwise stated.

Also, the term 'module' or 'unit' used in the specification means a software or hardware component, and the 'module' or 'unit' performs certain roles. However, 'module' or 'unit' is not meant to be limited to software or hardware. A 'module' or 'unit' may be configured to reside in an addressable storage medium and may be configured to reproduce one or more processors. Thus, as an example, a 'module' or 'unit' includes components such as software components, object-oriented software components, class components, and task components, processes, functions, and attributes. , procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, or variables. Functions provided within components and 'modules' or 'units' may be combined into a smaller number of components and 'modules' or 'units', or further components and 'modules' or 'units'. can be further separated.

According to one embodiment of the present disclosure, a 'module' or 'unit' may be implemented with a processor and a memory. 'Processor' should be interpreted broadly to include general-purpose processors, central processing units (CPUs), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some circumstances, 'processor' may refer to an application specific integrated circuit (ASIC), programmable logic device (PLD), field programmable gate array (FPGA), or the like. 'Processor' refers to a combination of processing devices, such as, for example, a combination of a DSP and a microprocessor, a combination of a plurality of microprocessors, a combination of one or more microprocessors in conjunction with a DSP core, or a combination of any other such configurations. You may. Also, 'memory' should be interpreted broadly to include any electronic component capable of storing electronic information. 'Memory' includes random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), programmable read-only memory (PROM), erasable-programmable read-only memory (EPROM), It may also refer to various types of processor-readable media, such as electrically erasable PROM (EEPROM), flash memory, magnetic or optical data storage, registers, and the like. A memory is said to be in electronic communication with the processor if the processor can read information from and/or write information to the memory. Memory integrated with the processor is in electronic communication with the processor.

In the present disclosure, a 'hyperscale language model' may refer to a language model capable of inference without fine-tuning using a method such as a few-shot learning, and It can have more than 10 times as many parameters as regular language models (e.g., more than 100 billion parameters, etc.). For example, the super large language model may be HyperCLOVA or Generative Pretrained Transformer 3 (GPT 3). In the present disclosure, a very large language model may be referred to as a language model.

In the present disclosure, a 'persona' may refer to a character that is distinguished from other characters or people by expressing unique linguistic characteristics due to a specific personality, a specific worldview, and a specific conversational style.

In the present disclosure, 'chatbot' may refer to artificial intelligence (AI)-based communication software that provides information related to a specific service or provides answers corresponding to questions included in a user's speech. A chatbot with a specific character persona may be referred to as a character chatbot.

In the present disclosure, 'speech' may refer to a verbal act of speaking out loud or a description of the corresponding verbal act in a text form (eg, text).

In the present disclosure, a 'conversation session' may refer to a conversation scene including a plurality of utterances in which two (or more than two) conversations participate. For example, a conversation session may be a conversation between a user and a character chatbot, and in this case, may include user speech and character speech.

1 is a diagram illustrating an example of constructing a final conversation database 160 including character conversations using a language model according to an embodiment of the present disclosure. The character chatbot may provide character speech maintaining a specific character persona in response to the user's speech. In order to create such a character chatbot, it is necessary to set the conversation style, world view, and personality of the character. Here, the dialogue style of the character may refer to the character's tone of voice. The tone of speech may refer to structural information of a sentence expressed independently of the meaning of an utterance. In addition, the character's view of the world may refer to a unique story possessed by the character. That is, the character chatbot may provide character utterances that maintain the context of the story constituting the background of the character. Also, the personality of a character may refer to a consistent conversation pattern of a character in a specific situation.

In order to create such a character chatbot including the dialogue style, world view, and personality of the character, a vast character dialogue database is required. However, it is difficult to build a conversation database because conversation data from SNS cannot be used due to problems such as leakage of personal information. A conversation database created using a super-large language model mainly creates conversations with a similar structure to the input sample conversations. There is a problem with In addition, there is a difficulty in that a lot of cost and effort are invested in inspecting the dialogue data that does not fit the character dialogue style, world view, and personality among the generated dialogue data.

In order to solve this problem, a dialogue database is created using a very large language model based on a plurality of dialogue purposes, dialogue types (actions), and dialogue situations, so that character dialogue data of various structures is stored. can create In addition, an efficient character conversation database can be constructed by filtering conversation data that does not match the dialogue style, world view, and personality of the character using the character persona detector 150 .

According to an embodiment, the dialogue database construction system constructs an initial dialogue database 130 by inputting a plurality of dialogue purpose-conversation type pairs 110 and a specific dialogue situation list 120 associated with each conversation purpose into a language model. can For example, the dialogue database building system constructs the initial dialogue database 130 by generating a plurality of seed dialogue sessions by inputting a plurality of dialogue purpose-conversation type-specific dialogue situation triplets into a very large language model. can

Then, the dialog database building system may build the augmented dialog database 140 by inputting each seed dialog session in the initial dialog database 130 into the super-large language model to generate at least one augmented dialog session. For example, the dialogue database construction system creates an augmented dialogue session by paraphrasing utterances included in each seed dialogue session in the initial dialogue database 130 using a very large language model, thereby generating the augmented dialogue database 140. ) can be built. Here, the initial dialogue database 130 and the augmented dialogue database 140 may refer to the same dialogue database. In this case, the augmented dialog database 140 may refer to a dialog database in which an enhanced dialog session is added to the initial dialog database 130 built by creating a seed dialog session.

After that, the dialogue database construction system inputs the plurality of seed dialogue sessions and the plurality of augmented dialogue sessions included in the augmented dialogue database 140 to the character persona detector 150 to filter dialogue sessions that do not match the character persona. there is. For example, when the character persona detector 150 determines that a specific chat session does not match the character of the character chatbot, the corresponding chat session may be removed. Also, when the character persona detector 150 determines that a specific augmented chat session is false, the corresponding chat session may be removed. In addition, the character persona detector 150 may convert character utterances in a specific conversation session into character conversations if they do not match the character conversations. The conversation database building system may create the final conversation database 160 with the filtered conversation sessions.

Through this configuration, even without collecting SNS conversation data, it is possible to easily and quickly create a vast amount of character conversation database for character chatbot creation using a super-large language model. In addition, it is possible to build a high-quality character conversation database by automatically inspecting the created conversation database without excessive human effort.

2 is a schematic diagram illustrating a configuration in which an information processing system 230 is connected to communicate with a plurality of user terminals 210_1, 210_2, and 210_3 in order to provide a character chatbot service according to an embodiment of the present disclosure. The information processing system 230 may include a system capable of providing a character chatbot service. In one embodiment, the information processing system 230 includes one or more server devices and/or databases capable of storing, providing, and executing computer-executable programs (eg, downloadable applications) and data related to the character chatbot service, or It may include one or more distributed computing devices and/or distributed databases based on cloud computing services. For example, the information processing system 230 may include separate systems (eg, servers) for character chatbot services. The character chatbot service provided by the information processing system 230 is provided to users through instant messaging applications, artificial intelligence-based communication software, web browsers, etc. installed in each of the plurality of user terminals 210_1, 210_2, and 210_3. can In one embodiment, the information processing system 230 may provide a character chatbot service to a user terminal using the super large language model 240 . Also, the information processing system 230 may build a conversation database for generating a character chatbot using the super large language model 240 .

A plurality of user terminals 210_1 , 210_2 , and 210_3 may communicate with the information processing system 230 through the network 220 . The network 220 may be configured to enable communication between the plurality of user terminals 210_1, 210_2, and 210_3 and the information processing system 230. Depending on the installation environment, the network 220 may be, for example, a wired network such as Ethernet, a wired home network (Power Line Communication), a telephone line communication device and RS-serial communication, a mobile communication network, a wireless LAN (WLAN), It may consist of a wireless network such as Wi-Fi, Bluetooth, and ZigBee, or a combination thereof. The communication method is not limited, and the user terminals (210_1, 210_2, 210_3) as well as a communication method utilizing a communication network (eg, mobile communication network, wired Internet, wireless Internet, broadcasting network, satellite network, etc.) that the network 220 may include ), short-range wireless communication between them may also be included.

In FIG. 2, a mobile phone terminal 210_1, a tablet terminal 210_2, and a PC terminal 210_3 are illustrated as examples of user terminals, but are not limited thereto, and the user terminals 210_1, 210_2, and 210_3 may perform wired and/or wireless communication. It can be any computing device capable of this. For example, the user terminal includes a smartphone, a mobile phone, a navigation device, a computer, a laptop computer, a digital broadcasting terminal, a personal digital assistant (PDA), a portable multimedia player (PMP), a tablet PC, a game console, and a wearable device ( wearable device), internet of things (IoT) device, virtual reality (VR) device, augmented reality (AR) device, and the like. In addition, although three user terminals 210_1, 210_2, and 210_3 are shown in FIG. 2 to communicate with the information processing system 230 through the network 220, it is not limited thereto, and a different number of user terminals may be connected to the network ( 220) to communicate with the information processing system 230.

In one embodiment, the information processing system 230 may receive user speech (eg, text, voice, etc.) from the user terminals 210_1, 210_2, and 210_3. In this case, the information processing system 230 may provide character speech maintaining the persona of the character chatbot responding to the received user speech information. Although the very large language model 240 is shown to exist outside the information processing system 230 in FIG. 2, it is not limited thereto, and the very large language model 240 may be stored and used inside the information processing system 230.

3 is a block diagram illustrating the internal configuration of the user terminal 210 and the information processing system 230 and the super large language model 240 according to an embodiment of the present disclosure. The user terminal 210 may refer to any computing device capable of executing an instant messaging application, artificial intelligence-based communication software, a web browser, etc. and capable of wired/wireless communication. For example, the mobile phone terminal 210_1 of FIG. 2 ), a tablet terminal 210_2, a PC terminal 210_3, and the like. As shown, the user terminal 210 may include a memory 312 , a processor 314 , a communication module 316 and an input/output interface 318 . Similarly, the information processing system 230 may include a memory 332 , a processor 334 , a communication module 336 and an input/output interface 338 . As shown in FIG. 3, the user terminal 210 and the information processing system 230 are configured to communicate information and/or data through the network 220 using respective communication modules 316 and 336. It can be. In addition, the input/output device 320 may be configured to input information and/or data to the user terminal 210 through the input/output interface 318 or output information and/or data generated from the user terminal 210.

The memories 312 and 332 may include any non-transitory computer readable media. According to one embodiment, the memories 312 and 332 are non-perishable mass storage devices such as random access memory (RAM), read only memory (ROM), disk drives, solid state drives (SSDs), flash memory, and the like. (permanent mass storage device) may be included. As another example, a non-perishable mass storage device such as a ROM, SSD, flash memory, or disk drive may be included in the user terminal 210 or the information processing system 230 as a separate permanent storage device separate from memory. Also, an operating system and at least one program code may be stored in the memories 312 and 332 .

These software components may be loaded from a computer readable recording medium separate from the memories 312 and 332 . A recording medium readable by such a separate computer may include a recording medium directly connectable to the user terminal 210 and the information processing system 230, for example, a floppy drive, a disk, a tape, a DVD/CD- It may include a computer-readable recording medium such as a ROM drive and a memory card. As another example, software components may be loaded into the memories 312 and 332 through the communication modules 316 and 336 rather than computer-readable recording media. For example, at least one program is loaded into the memories 312 and 332 based on a computer program installed by files provided by developers or a file distribution system that distributes application installation files through the network 220 . It can be.

The processors 314 and 334 may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input/output operations. Instructions may be provided to processors 314 and 334 by memories 312 and 332 or communication modules 316 and 336 . For example, processors 314 and 334 may be configured to execute instructions received according to program codes stored in a recording device such as memory 312 and 332 .

The communication modules 316 and 336 may provide configurations or functions for the user terminal 210 and the information processing system 230 to communicate with each other through the network 220, and the user terminal 210 and/or information processing. System 230 may provide configurations or functions for communicating with other user terminals or other systems (eg, separate cloud systems, etc.). For example, a request or data generated by the processor 314 of the user terminal 210 according to a program code stored in a recording device such as the memory 312 (eg, data including user speech such as text or voice) may be delivered to the information processing system 230 through the network 220 under the control of the communication module 316. Conversely, a control signal or command provided under the control of the processor 334 of the information processing system 230 passes through the communication module 336 and the network 220 and through the communication module 316 of the user terminal 210. It may be received by the user terminal 210 . For example, the user terminal 210 may receive character speech in response to user speech from the information processing system 230 .

The input/output interface 318 may be a means for interfacing with the input/output device 320 . As an example, the input device may include a device such as a camera, keyboard, microphone, mouse, etc. including an audio sensor and/or image sensor, and the output device may include a device such as a display, speaker, haptic feedback device, or the like. can As another example, the input/output interface 318 may be a means for interface with a device in which a configuration or function for performing input and output is integrated into one, such as a touch screen. Although the input/output device 320 is not included in the user terminal 210 in FIG. 3 , it is not limited thereto, and the user terminal 210 and the user terminal 210 may be configured as one device. In addition, the input/output interface 338 of the information processing system 230 is connected to the information processing system 230 or means for interface with a device (not shown) for input or output that the information processing system 230 may include. can be In FIG. 3, the input/output interfaces 318 and 338 are shown as separate elements from the processors 314 and 334, but are not limited thereto, and the input/output interfaces 318 and 338 may be included in the processors 314 and 334. there is.

The user terminal 210 and the information processing system 230 may include more components than those shown in FIG. 3 . However, there is no need to clearly show most of the prior art components. In one embodiment, the user terminal 210 may be implemented to include at least some of the aforementioned input/output devices 320 . In addition, the user terminal 210 may further include other components such as a transceiver, a global positioning system (GPS) module, a camera, various sensors, and a database. For example, when the user terminal 210 is a smartphone, it may include components that are generally included in a smartphone, for example, an acceleration sensor, a gyro sensor, a microphone module, a camera module, and various physical Various components such as a button, a button using a touch panel, an input/output port, and a vibrator for vibration may be implemented to be further included in the user terminal 210 .

4 is a diagram illustrating an example of generating a plurality of conversation purpose-conversation type pairs 110 according to an embodiment. According to one embodiment, the dialogue purpose-dialogue type pair 110 may be used to define the dialogue purpose and dialogue type of a character utterance responding to a user utterance. A conversation purpose and conversation type of character speech may be predefined by a character chatbot service provider.

The conversation purpose may represent a character's attitude in conversation with the user. For example, as shown, conversational purposes may be defined as anger 410, comfort 420, and the like. That is, with respect to the user's utterance, the character may comfort a farewell or congratulate a college passer. Conversational purpose can be an important factor in determining the flow of a conversational session. Although anger 410 and consolation 420 are shown as examples of conversation purposes in FIG. 4 , it is not limited thereto, and any number of conversation purposes may be defined and created.

The dialogue type may indicate a pattern of character speech in response to user speech. For example, as shown, the conversation type includes a conversation type (412, 422) in which the character chatbot sympathizes with the user's utterance, a conversation type (414, 424) providing feedback, and a conversation type (416) sharing experiences. , 426) and the like. In this way, a deeper conversation structure can be implemented by defining various ways in which characters can participate in conversation as various conversation types.

Also, a conversation purpose and a conversation type may be defined in association with each other. For example, as shown, a conversation type 412 that the character chatbot empathizes with in relation to a conversation purpose 410 in which the character chatbot is angry in response to a user utterance, a conversation type 414 that provides feedback, and experiences A sharing conversation type 416 may be defined. Similarly, a conversation type 422 that the character chatbot empathizes with in connection with the conversation purpose 420 that the character chatbot comforts in response to user utterance, a conversation type that provides feedback 424, and a conversation type that shares experiences ( 426) can be defined. In one embodiment, a pair of conversation purpose and conversation type may be defined according to the personality of the character to be created. That is, it is possible to control the character of the character according to the predetermined conversation purpose and conversation type. That is, in the same conversation situation, a specific character gives sympathy while being angry, but another character shares experiences while being angry. Another character can empathize with comfort instead of anger in the same situation.

5 is a diagram illustrating an example of generating a schematic conversation situation 520 according to an exemplary embodiment. According to an embodiment, the dialog database construction system may generate a dialog situation list using the super large language model 240 . Here, the conversation situation means a context and situation in which a conversation occurs, and the conversation situation list may include a plurality of specific conversation situations (eg, 120 in FIG. 1 ) and a plurality of schematic conversation situations 520 . A specific conversation situation (eg, 120 in FIG. 1 ) may be generated based on a general conversation situation. In one embodiment, the general conversation situation and the specific conversation situation may be referred to as a first-level conversation situation and a second-level conversation situation, respectively. Here, the dialogue situation of the second level may be a dialogue situation of a lower level that describes a more specific dialogue situation than the first level dialogue situation.

According to one embodiment, the conversation database building system provides a prompt 510 including a first conversation objective 512 , a plurality of schematic conversation situation samples 514 associated with the first conversation objective 514 and a second conversation objective 516 . may be input into the super large language model 240 to generate a plurality of schematic conversation situations 520 associated with the second conversation purpose. In one embodiment, the first conversational purpose 512 and the second conversational purpose 516 may be written in the form of directives. For example, the first conversation purpose 512 may be included in the prompt 510 in the form of a directive such as 'specific situations that cause anger together'. Similarly, the second conversation purpose 516 may be included in the prompt 510 in the form of a directive such as 'specific warm comforting situations'. A plurality of schematic conversation situation samples 514 associated with the first conversation purpose may be predefined by a service provider.

The dialog database construction system inputs a prompt including another dialog purpose (eg, congratulations, etc.) into the extra large language model 240 instead of the second dialog purpose 516, thereby generating a plurality of schematic conversation situations related to the other conversation purpose. can create When several conversation purposes are used to build a conversation database, the conversation database building system can generate a plurality of schematic conversation situations associated with each conversation purpose by repeating the above process.

6 is a diagram illustrating an example of generating a specific conversation situation 630 according to an embodiment. According to an embodiment, the dialog database construction system may generate various specific dialog situations related to a dialog purpose by using the super-large language model 240 . Specifically, the conversation database building system includes a first conversation purpose 612, a first schematic conversation situation 614 associated with the first conversation purpose, and a specific conversation situation sample 616 associated with the first conversation purpose (or the first schematic conversation situation 614). A prompt 610 including a specific dialogue situation sample associated with the dialogue situation 614), a second dialogue objective 618, and a second schematic dialogue situation 620 associated with the second dialogue objective ) to generate a specific conversation situation 630 associated with the second conversation purpose (or a specific conversation situation associated with the second general conversation situation).

In one embodiment, the first conversational purpose 612 and the second conversational purpose 618 may be written in the form of directives. For example, the first conversation purpose 612 and the second conversation purpose 618 may be included in the prompt 610 in the form of directives such as 'contents to be comforted'. A specific conversation situation sample 616 associated with the first conversation purpose may be predefined by a service provider.

The dialogue database construction system replaces the second schematic dialogue situation 620 with other schematic dialogue situations (eg, 'I got scolded at work', 'I was scolded by my parents' included in the schematic dialogue situation 520 of FIG. 5 ). A plurality of specific conversation situations related to other general conversation situations may be generated by inputting a prompt including 'uh' into the super-large language model 240 . When various general conversation situations are generated by the super-large language model 240, the conversation database building system may generate a plurality of specific conversation situations related to each general conversation situation by repeating the above-described process.

Although the first conversation purpose 612 and the second conversation purpose 618 are illustrated in FIG. 6 as the same, it is not limited thereto, and the first conversation purpose 612 and the second conversation purpose 618 may be different from each other. . As shown in FIGS. 5 and 6 , diversity can be secured primarily by generating a schematic dialogue situation first, and concreteness can be secured secondarily by creating a specific dialogue situation using a schematic dialogue situation. With this configuration, it is possible to densely secure each situation while expanding the overall situation coverage of the dialogue situation. Since this process is automatically performed using the super large language model 240, it can be easily extended. Through this configuration, even if the conversation between the user and the character chatbot deals with a specific topic, it is possible to have a customized conversation with the user according to the situation by providing high-quality speech of the character chatbot corresponding to the user's speech.

7 is a diagram illustrating an example of a prompt 700 for creating a seed conversation session, according to one embodiment. According to an embodiment, the dialogue database construction system constructs an initial dialogue database by generating a plurality of seed dialogue sessions corresponding to a plurality of dialogue purpose-conversation type-specific dialogue situation triplets using a super-large language model. can

Specifically, the dialogue database construction system includes a plurality of dialogue purpose-conversation type pairs (eg, 110 in FIG. 1 ) and a plurality of specific dialogue situations associated with each conversation purpose (eg, any one of 120 in FIG. 1 or 630 in FIG. 6 ). A plurality of conversation purpose-conversation type-specific conversation situation triple pairs may be generated based on the specific conversation situation). Then, the conversation database building system uses a first triple pair 710 (eg, a first conversation purpose-conversation type-specific conversation situation) as sample data for generating a seed conversation session, and a conversation session corresponding to the first triple pair. A prompt 700 containing a sample 720 and a second triplet 730 (e.g., second conversation purpose-conversation type-specific conversation situation) as data associated with the seed conversation session to be created is input to the extra large language model , may create a seed conversation session associated with the second triplet. Here, the chat session sample 720 corresponding to the first triple pair includes at least one user utterance and at least one character utterance, and may be predefined by a service provider. A seed conversation session created by inputting the prompt 700 into the super-large language model may similarly include at least one user utterance and at least one character utterance.

In one embodiment, the prompt 700 may include a first triple pair 710 , a conversation session sample 720 corresponding to the first triple pair and a second triple pair 730 . For example, as shown, the first triple pair 710 includes a conversation purpose (eg, 'angry'), a conversation type (eg, 'to share one's own experience') and a specific conversation situation related thereto (eg, 'outrage'). , 'I came out without an umbrella because it wasn't raining and I got wet'). In addition, the conversation session sample 720 corresponding to the first triple pair includes a speech sample as a speech of a character chatbot corresponding to the first triple pair (eg, 'I didn't have an umbrella last time, so I got hit in the rain! ㅠ') can do. The second triplet 730 may include a conversation purpose (eg, 'angry'), a conversation type (eg, 'introducing and sharing experiences'), and a specific conversation context associated with the seed conversation session to be created. In addition, the prompt 700 may be predefined by a character chatbot service provider, and as many as the number of predefined conversation purpose-conversation type pairs (eg, 110 in FIG. 1 ) may be predefined by the service provider. The dialog database construction system inputs a plurality of specific conversation situations related to the corresponding conversation purpose (eg, 'anger') one by one in the second triple pair 730 position in each prompt 700, and You can create seed chat sessions.

8 is a diagram illustrating an example of generating an augmented chat session 820 corresponding to a seed chat session 810 using a super large language model 240 according to an embodiment of the present disclosure. According to one embodiment, the dialog database building system may build the augmented dialog database by using the super large language model 240 to create an augmented dialog session 820 for each seed dialog session in the initial dialog database. For example, as shown, the dialog database building system generates at least one augmented dialog session 820 by inputting a seed dialog session 810 including user utterances and character utterances into a super large language model 240. can do. The augmented dialog session 820 may similarly include user utterances and character utterances. Here, the augmented dialog database may refer to a database in which an enhanced dialog session is added to an initial dialog database.

In one embodiment, the dialog database building system may paraphrase the utterances within the seed dialog session 810 using the very large language model 240 to create the augmented dialog session 820 . As shown, user utterances in the seed conversation session 810 (e.g., 'The uncle next door is going to die because he smokes without even trying') has the same meaning and context, but has a different expression method such as vocabulary (user utterances) For example, it can be paraphrased as 'My neighbor keeps smoking, so the smell keeps coming into my house'). Similarly, character utterances within the seed conversation session 810 (e.g., 'Oops! That's such a nuisance! Why are you hurting others? Go and protest right away!') have the same meaning and context, but the expression method such as vocabulary This can be paraphrased with other character utterances (e.g., 'Oops, that's too much! I'll go pick it up right away!').

8 shows that both the user utterance and the character utterance included in the seed chat session 810 are paraphrased to create the augmented chat session 820, but it is not limited thereto, and only the user utterance or character utterance is paraphrased. An enhanced conversation session 820 may be created. 8 shows an example in which one augmented chat session 820 is generated for one seed chat session 810, but is not limited thereto, and multiple augmentations are generated for one seed chat session 810. A chat session may be created. In this case, a plurality of intermediary chat sessions associated with one seed chat session 810 may have the same meaning and context, but may have different expressions such as vocabulary.

9 is a diagram illustrating examples of predefined character personality maps 910 and 920 according to an embodiment of the present disclosure. To create a character chatbot, it is necessary to set the character's personality. Here, the personality of a character may be defined as a consistent conversation pattern of a character in a specific situation. A specific situation may correspond to a conversation purpose, and a consistent conversation pattern of a character may correspond to a conversation type. That is, the personality of the character can be expressed as a set of conversation purpose-conversation type pairs, and the set of conversation purpose-conversation type pairs can be expressed as character personality maps 910 and 920 .

For example, as shown, the first character personality map 910 includes 'anger'-'share experience', 'congratulation'-'share experience', 'anger'-'sympathy', and 'congratulation'-'sympathy'. It can be set as a set of conversation purpose-conversation type pairs such as '. In this case, the character of the corresponding character has a conversation pattern of sharing experiences and empathizing in situations of 'anger' and 'congratulation', so it can be defined as a personality that reacts well and talks a lot about oneself. Similarly, the second character personality map 920 may be set as a set of conversation purpose-conversation type pairs such as 'up'-'feedback' or 'up'-'face-to-face'. In this case, the character of the character can be defined as a hot-tempered character who is good at making direct remarks, since he has a conversational pattern of providing immediate feedback and facing the situation in a 'comfort' situation.

10 is a diagram illustrating an example of generating a final conversation database 160 using a character persona detector 150 according to an embodiment of the present disclosure. According to an embodiment, the dialogue database construction system may construct the initial dialogue database 130 by inputting a plurality of dialogue purpose-conversation type pairs and a detailed dialogue situation list associated with each dialogue purpose to a super-large language model. Here, a plurality of conversation purpose-conversation type pairs may be defined based on a predefined character personality map 1010 . Alternatively, the plurality of conversation purpose-conversation type pairs may include all combinations of conversation purpose-conversation type pairs. Then, the dialog database building system may build the augmented dialog database 140 by inputting each seed dialog session in the initial dialog database 130 into the super-large language model to generate at least one augmented dialog session.

Then, the conversation database building system inputs the plurality of seed conversation sessions and the plurality of augmented conversation sessions included in the augmented conversation database 140 to the character persona detector 150 to filter out conversation sessions that do not match the character persona, and finally Conversation database 160 can be built. As shown, the character persona detector 150 may include a personality filter 1020, a worldview filter 1030, a conversational filter 1040, and a conversational changer 1050. In one embodiment, character persona detector 150 may input seed conversation sessions and/or augmented conversation sessions into personality filter 1020 to remove augmented conversation sessions that do not match a predefined character personality map. Additionally or alternatively, the character persona detector 150 inputs the augmented dialog session into the personality filter 1020 to generate a seed dialog session whose estimated conversation purpose-conversation type is associated with the augmented conversation session. If it does not match the conversation purpose-conversation type used at the time, the corresponding augmented conversation session may be removed.

Thereafter, the character persona detector 150 may use the worldview filter 1030 to verify authenticity of the augmented chat session and remove the augmented chat session determined to be false. Then, the character persona detector 150 may use the dialog filter 1040 to identify character utterances that do not match the character dialog among character utterances included in the seed conversation session and/or the augmented chat session. If a character utterance that does not match the character dialogue is identified, the character persona detector 150 may use the dialogue converter 1050 to convert the character utterance that does not match the character dialogue into a character dialogue. The dialogue database construction system may build the final dialogue database 160 using the filtered dialogue sessions and learn/create the character chatbot 1060 using a plurality of dialogue sessions included in the final dialogue database 160. .

In one embodiment, personality filter 1020, worldview filter 1030, dialog filter 1040 and dialog converter 1050 are the result of a service provider manually reviewing some of the seed dialog sessions in initial dialog database 130. It may be a machine-learned model using Through this configuration, it is possible to provide the character chatbot 1060 in which the personality of the character is well revealed to the user by easily and quickly generating the persona conversation data.

11 is a diagram illustrating an example of filtering a conversation session 1110 using a personality filter 1020 according to an embodiment of the present disclosure. Here, chat session 1110 may be a seed chat session or an enhanced chat session in an enhanced chat database. According to an embodiment, the conversation database system may input the conversation session 1110 to the personality filter 1020 and remove the conversation session if it does not match a predefined character personality map.

Specifically, the conversation database system may input the conversation session 1110 to the personality filter 1020 to estimate the conversation purpose 1120 and the conversation type 1130 of the conversation session 1110 . The conversation database system can then determine whether the inferred conversation purpose 1120 and conversation type 1130 match a predefined character personality map. If it determines that the estimated conversation purpose 1120 and conversation type 1130 do not match the predefined character personality map, the conversation database system may remove the conversation session 1110 . On the other hand, if it is determined that the estimated conversation purpose 1120 and conversation type 1130 match the predefined character personality map, the conversation database system may maintain the conversation session 1110 .

In FIG. 11 , the personality filter 1020 is illustrated as removing conversation sessions that do not match a predefined character personality map, but is not limited thereto. For example, the conversation purpose 1120 and conversation type 1130 of the augmented conversation session 1110 entered by the personality filter 1020 are estimated, and the estimated conversation purpose 1120 and conversation type 1130 are the augmented conversation Enhanced chat session 1120 may be removed if it does not match the conversation purpose and conversation type used when creating the seed conversation session associated with session 1110 . With this configuration, the conversation database system can create a persona conversation database with higher quality by removing conversation sessions that do not match the character's personality.

12 is a diagram illustrating an example of filtering an augmented conversation session 1220 using a worldview filter 1030 according to an embodiment of the present disclosure. According to one embodiment, the conversation database system may input seed conversation sessions 1210 and corresponding enhanced conversation sessions 1220 to worldview filter 1030 to remove augmented conversation sessions 1220 determined to be false. Here, the augmented chat session 1220 may be created by paraphrasing the seed chat session 1210 .

In one embodiment, the conversation database system may input seed conversation session 1210 and augmented conversation session 1220 into worldview filter 1030 to verify authenticity of augmented conversation session 1220 . For example, a dialog database system may use a Natural Language Inference (NLI) model to determine a relationship between a seed dialog session 1210 and an augmented dialog session 1220 associated with the seed dialog session 1210. . Here, the NLI model may be a model for determining whether one sentence is logically related to another sentence, and the relationship between the seed conversation session 1210 and the augmented conversation session 1220 associated with the seed conversation session 1210 is neutral. It can be determined by classifying it into a neutral relationship, an entailment relationship, and a contradiction relationship. If it determines that the seed conversation session 1210 and the augmented conversation session 1220 associated with the seed conversation session 1210 are not synonymy, the worldview filter 1030 may determine the enhanced conversation session to be false.

In one embodiment, if the enhanced chat session 1220 is determined to be false, the conversation database system may remove the enhanced chat session 1220 . On the other hand, if it is determined that the enhanced chat session 1220 is not fake, the chat database system may maintain the corresponding enhanced chat session 1220 . With this configuration, the conversation database system can create a persona conversation database with higher quality by removing augmented conversation sessions whose meanings are unintentionally changed in the paraphrasing process.

FIG. 13 is a diagram illustrating an example of converting character utterances that do not match a character dialogue into a character dialogue using a dialogue filter 1040 and a dialogue converter 1050 according to an embodiment of the present disclosure. Here, chat session 1110 may be a seed chat session or an enhanced chat session in an enhanced chat database. As shown, the dialogue database system may input the dialogue session 1310 to the dialogue filter 1040 to identify character speech that does not match the character dialogue among character speech included in the dialogue session 1310 . For example, the conversational filter 1040 is the first character utterance 1312 'huh? oh my god! It can be judged that 'Oh, I'm bored, let it go~' matches the character dialogue, and the second character utterance (1314), 'I believe that human efforts are definitely not betrayed' does not match the character dialogue.

In one embodiment, the dialogue database system may input the second character speech 1314 identified as a character speech that does not match the character dialogue dialogue to the character dialogue converter 1050 and convert it into the character dialogue dialogue 1320. For example, the dialogue converter 1050 converts the second character speech 1314 that does not match the character dialogue into the third character speech 1320 having the character dialogue 'Effort never betrays!' can be converted In this case, the conversation database system may replace and store the second character utterance 1314 with the third character utterance 1320 . With this configuration, the dialogue database system can create a persona dialogue database with higher quality by correcting dialogue sessions that do not match the character dialogue.

14 is a flowchart illustrating a method 1400 of constructing a conversation database using a language model according to an embodiment of the present disclosure. In one embodiment, the conversation database construction method 1400 may be performed by a processor (eg, a user terminal or at least one processor of an information processing system). In another embodiment, the information processing system and the user terminal may separately perform the steps of the method 1400 for constructing a conversation database.

In one embodiment, the conversation database construction method 1400 may be initiated by the processor receiving a plurality of predefined conversation purpose-conversation type pairs (S1410). Then, the processor may generate a plurality of conversation situations associated with each conversation purpose by using the language model (S1420). To this end, the processor generates a plurality of schematic dialog situations associated with the second dialog purpose by inputting a prompt including a first dialog purpose, a schematic dialog situation sample associated with the first dialog purpose, and a second dialog purpose to the language model. can do. Here, the plurality of general conversation situations associated with the second conversation purpose may include a first general conversation situation and a second general conversation situation. And, the processor generates a set of first specific dialogue situations associated with the second dialogue objective by inputting a prompt including a second dialogue objective, a specific dialogue situation sample associated with the second dialogue objective, and a first schematic dialogue situation to the language model. can do. Similarly, the processor inputs a prompt including a second dialog objective, a sample of concrete dialog situations associated with the second dialog objective, and a second high-level dialog situation to the language model to obtain a second set of specific dialog situations associated with the second dialog objective. can create

After all dialogue situations are generated, the processor may build a plurality of seed conversation sessions corresponding to a plurality of conversation purpose-conversation type-conversation situation triplets using a language model to build an initial conversation database ( S1430). Here, each seed conversation session may include at least one user utterance and at least one character utterance. Specifically, the processor may generate a plurality of conversation purpose-conversation type-conversation situation triplets based on a plurality of conversation purpose-conversation type pairs and a plurality of conversation situations associated with each conversation purpose. And, the processor may provide a prompt including a first conversation purpose-conversation type-conversation situation, a conversation session sample corresponding to the first conversation purpose-conversation type-conversation situation, and a second conversation purpose-conversation type-conversation situation to the language model. A seed conversation session associated with the second conversation purpose-conversation type-conversation situation may be created by inputting the seed conversation session. Here, the conversation session sample corresponding to the first conversation purpose, conversation type, and conversation situation may include at least one user utterance and at least one character utterance.

In one embodiment, the processor may use the language model to create and add at least one augmented conversation session to the conversation database for each conversation session in the initial conversation database. Here, each augmented dialog session may include utterances paraphrasing utterances in an associated seed dialog session. Thereafter, the processor may input the plurality of conversation sessions included in the conversation database to the character persona detector to filter conversation sessions that do not match the character persona to generate a final conversation database.

In one embodiment, the processor inputs a first enhanced dialog session included in a plurality of enhanced dialog sessions to a personality filter to estimate a conversation purpose and conversation type, and the estimated conversation purpose and estimated conversation type of the first enhanced conversation session. In response to determining that it does not match this predefined character personality map, the first augmented conversation session may be removed. Alternatively or additionally, the processor inputs the second enhanced dialog session included in the plurality of enhanced dialog sessions to the personality filter to estimate the conversation purpose and the conversation type, and the estimated conversation purpose and the estimated conversation of the second enhanced conversation session. In response to determining that the type does not match the conversation purpose and conversation type used in creating the seed conversation session associated with the second enhanced conversation session, the second enhanced conversation session may be removed.

Alternatively or additionally, the processor may input a third enhanced dialog session included in the plurality of enhanced dialog sessions and a seed dialog session associated with the third enhanced dialog session into a worldview filter to verify authenticity of the third enhanced dialog session. there is. In response to the third enhanced dialog session being determined to be false, the processor can remove the third enhanced dialog session. For example, the processor determines a relationship between the third enhanced dialog session and a seed dialog session associated with the third enhanced dialog session using a Natural Language Inference (NLI) model, and In response to determining that the seed conversation session associated with the enhanced conversation session is not an entailment, authenticity of the third enhanced conversation session may be verified by determining that the third enhanced conversation session is false.

Alternatively or additionally, the process may input a specific dialogue session of the plurality of dialogue sessions to a dialogue filter to identify character utterances included in the specific dialogue session that do not match a predefined character dialogue. Then, the processor may convert the identified character utterance into a predefined character dialog by inputting it to a character dialog converter.

In one embodiment, the processor may use the final conversation database to learn a character chatbot having a character persona.

The above method may be provided as a computer program stored in a computer readable recording medium to be executed on a computer. The medium may continuously store programs executable by a computer or temporarily store them for execution or download. In addition, the medium may be various recording means or storage means in the form of a single or combined hardware, but is not limited to a medium directly connected to a certain computer system, and may be distributed on a network. Examples of the medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROM and DVD, magneto optical media such as floptical disks, and Anything configured to store program instructions may include a ROM, RAM, flash memory, or the like. In addition, examples of other media include recording media or storage media managed by an app store that distributes applications, a site that supplies or distributes various other software, and a server.

The methods, acts or techniques of this disclosure may be implemented by various means. For example, these techniques may be implemented in hardware, firmware, software, or combinations thereof. Those skilled in the art will appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchange of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the particular application and the design requirements imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementations should not be interpreted as causing a departure from the scope of the present disclosure.

In a hardware implementation, the processing units used to perform the techniques may include one or more ASICs, DSPs, digital signal processing devices (DSPDs), programmable logic devices (PLDs) ), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, electronic devices, and other electronic units designed to perform the functions described in this disclosure. , a computer, or a combination thereof.

Accordingly, the various illustrative logical blocks, modules, and circuits described in connection with this disclosure may be incorporated into a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or may be implemented or performed in any combination of those designed to perform the functions described in A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, eg, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other configuration.

In firmware and/or software implementation, the techniques include random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), PROM ( on a computer readable medium, such as programmable read-only memory (EPROM), erasable programmable read-only memory (EPROM), electrically erasable PROM (EEPROM), flash memory, compact disc (CD), magnetic or optical data storage device, or the like. It can also be implemented as stored instructions. Instructions may be executable by one or more processors and may cause the processor(s) to perform certain aspects of the functionality described in this disclosure.

If implemented in software, the techniques may be stored on or transmitted over as one or more instructions or code on a computer readable medium. Computer readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a computer. By way of non-limiting example, such computer readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or desired program code in the form of instructions or data structures. It can be used for transport or storage to and can include any other medium that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium.

For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, the coaxial cable , fiber optic cable, twisted pair, digital subscriber line, or wireless technologies such as infrared, radio, and microwave are included within the definition of medium. Disk and disc as used herein include CD, laser disc, optical disc, digital versatile disc (DVD), floppy disk, and blu-ray disc, where disks are usually magnetic data is reproduced optically, whereas discs reproduce data optically using a laser. Combinations of the above should also be included within the scope of computer readable media.

A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium can be coupled to the processor such that the processor can read information from or write information to the storage medium. Alternatively, the storage medium may be integral to the processor. The processor and storage medium may reside within an ASIC. An ASIC may exist within a user terminal. Alternatively, the processor and storage medium may exist as separate components in a user terminal.

Although the embodiments described above have been described as utilizing aspects of the presently-disclosed subject matter in one or more stand-alone computer systems, the disclosure is not limited thereto and may be implemented in conjunction with any computing environment, such as a network or distributed computing environment. . Further, aspects of the subject matter in this disclosure may be implemented in a plurality of processing chips or devices, and storage may be similarly affected across multiple devices. These devices may include PCs, network servers, and portable devices.

Although the present disclosure has been described in relation to some embodiments in this specification, various modifications and changes may be made without departing from the scope of the present disclosure that can be understood by those skilled in the art. Moreover, such modifications and variations are intended to fall within the scope of the claims appended hereto.

110: multiple conversation purpose-dialogue type pairs
120: list of specific conversation situations
130 Initial Conversation Database
140 augmented dialog database
150: character persona detector
160: final conversation database

Claims (16)

A method for constructing a dialogue database using a language model, performed by at least one processor, comprising:
receiving a plurality of predefined conversation purpose-conversation type pairs;
generating a plurality of conversation situations associated with each conversation purpose by using a language model; and
constructing an initial conversation database by creating a plurality of seed conversation sessions corresponding to a plurality of conversation purpose-conversation type-conversation situation triplets using the language model;
including,
A method of building a conversation database, wherein each seed conversation session includes at least one user utterance and at least one character utterance. According to claim 1,
generating and adding at least one augmented dialog session to a dialog database for each dialog session in the initial dialog database using the language model;
Further comprising, a conversation database construction method. According to claim 2,
A method of building a dialog database, wherein each augmented dialog session includes utterances that are paraphrasing of utterances in an associated seed dialog session. According to claim 1,
The step of generating a plurality of conversation situations associated with each conversation purpose,
generating a plurality of schematic dialogue situations associated with the second dialogue purpose by inputting a prompt including a first dialogue purpose, a sample of a schematic dialogue situation associated with the first dialogue purpose, and a second dialogue purpose into the language model;
including,
The dialog database construction method of claim 1, wherein the plurality of general dialog situations associated with the second dialog purpose includes a first general dialog situation and a second general dialog situation. According to claim 4,
The step of generating a plurality of conversation situations associated with each conversation purpose,
A prompt including the second dialogue objective, a specific dialogue situation sample associated with the second dialogue purpose, and the first general dialogue situation is input to the language model to obtain a first specific dialogue situation set associated with the second dialogue purpose. generating; and
A prompt including the second dialogue objective, a specific dialogue situation sample associated with the second dialogue purpose, and the second general dialogue situation is input to the language model to obtain a second set of specific dialogue situations associated with the second dialogue purpose. steps to create
Further comprising, a conversation database construction method. According to claim 1,
The step of building the initial conversation database,
generating a plurality of conversation purpose-conversation type-conversation situation triplets based on the plurality of conversation purpose-conversation type pairs and a plurality of conversation situations associated with each conversation purpose; and
A prompt including a first conversation purpose-conversation type-conversation situation, a conversation session sample corresponding to the first conversation purpose-conversation type-conversation situation, and a second conversation purpose-conversation type-conversation situation is input into the language model, creating a seed conversation session associated with the second conversation purpose-conversation type-conversation situation;
including,
wherein the conversation session sample corresponding to the first conversation purpose-dialogue type-dialogue situation includes at least one user utterance and at least one character utterance. According to claim 2,
generating a final conversation database by inputting a plurality of conversation sessions included in the conversation database into a character persona detector and filtering conversation sessions that do not match the character persona;
Further comprising, a conversation database construction method. According to claim 7,
The step of creating the final conversation database,
estimating a conversation purpose and a conversation type by inputting a first augmented conversation session included in the plurality of augmented conversation sessions to a personality filter; and
In response to determining that the estimated conversation purpose and the estimated conversation type of the first enhanced conversation session do not match a predefined character personality map, removing the first enhanced conversation session.
Including, Conversation database construction method. According to claim 7,
The step of creating the final conversation database,
estimating a conversation purpose and a conversation type by inputting a second augmented conversation session included in the plurality of enhanced conversation sessions to a personality filter; and
In response to determining that an estimated conversation purpose and an estimated conversation type of the second enhanced conversation session do not match a conversation purpose and conversation type used when creating a seed conversation session associated with the second enhanced conversation session, removing the second enhanced dialog session;
Including, Conversation database construction method. According to claim 7,
The step of creating the final conversation database,
verifying authenticity of the third augmented chat session by inputting a third augmented chat session included in the plurality of augmented chat sessions and a seed chat session associated with the third augmented chat session into a world view filter; and
In response to the third enhanced dialog session being determined to be false, removing the third enhanced dialog session.
Including, Conversation database construction method. According to claim 10,
The step of verifying whether the third augmented conversation session is authentic,
determining a relationship between the third enhanced dialog session and a seed dialog session associated with the third enhanced dialog session using a Natural Language Inference (NLI) model; and
In response to determining that the third enhanced dialog session and a seed dialog session associated with the third enhanced dialog session are not entailments, determining the third enhanced dialog session to be false.
Including, Conversation database construction method. According to claim 7,
The step of creating the final conversation database,
inputting a specific dialogue session among the plurality of dialogue sessions to a dialogue filter to identify character speech that does not match a predefined character dialogue among character speech included in the specific dialogue session;
Including, Conversation database construction method. According to claim 12,
The step of creating the final conversation database,
Converting the identified character utterance into the predefined character dialogue by inputting it to a character dialogue converter
Further comprising, a conversation database construction method. According to claim 7,
Learning a character chatbot having a character persona using the final conversation database
Further comprising, a conversation database construction method. A computer program stored in a computer-readable recording medium to execute the method according to any one of claims 1 to 14 on a computer. As a dialogue database construction system,
communication module;
Memory; and
at least one processor connected to the memory and configured to execute at least one computer readable program contained in the memory
including,
The at least one program,
receive a plurality of predefined conversation purpose-conversation type pairs;
Creating a plurality of conversation situations related to each conversation purpose using a language model;
Including instructions for constructing an initial conversation database by creating a plurality of seed conversation sessions corresponding to a plurality of conversation purpose-conversation type-conversation situation triplets using the language model;
Wherein each seed dialogue session includes at least one user utterance and at least one character utterance.

Images