WO2019156536A1

WO2019156536A1 - Method and computer device for constructing or updating knowledge base model for interactive ai agent system by labeling identifiable, yet non-learnable, data from among learning data, and computer-readable recording medium

Info

Publication number: WO2019156536A1
Application number: PCT/KR2019/001693
Authority: WO
Inventors: 설재호; 장세영
Original assignee: 주식회사 머니브레인
Priority date: 2018-02-12
Filing date: 2019-02-12
Publication date: 2019-08-15
Also published as: KR101950387B1

Abstract

A method performed by a computer device is disclosed. The method is for automatically constructing or updating a knowledge base model for an interactive AI agent system and comprises the steps of: receiving a series of dialogue logs related to each other; identifying, from each of the received dialogue logs, entity information determined as requiring non-identification according to a predetermined criterion, wherein the entity information includes an entity type and a value; replacing, with each corresponding label, the value of respective pieces of identified entity information, wherein the label is an identifier for identifying an entity type of corresponding entity information, and pieces of entity information having the same type and the same value, which are included in the dialogue logs, are replaced with the same label; and constructing or updating the knowledge base model for the interactive AI agent system according to learning by using dialogue logs in which the value of the identified entity information is replaced with each corresponding label.

Description

Method of constructing or updating knowledge base model for interactive AI agent system, computer device, and computer readable recording medium through labeling of identifiable but non-learnable data in training data

The present disclosure relates to learning of an interactive AI agent system, and more particularly, to a method of processing learning data for an interactive AI agent system and learning the interactive AI agent system using the processed learning data. will be.

In recent years, with advances in the field of artificial intelligence, in particular in the field of understanding natural language, the user has moved away from the operation of the traditional machine-oriented command input / output schemes, thereby allowing users to use natural language in a more human-friendly manner, such as speech and / or text. Increasingly, the development and use of interactive AI agent systems, which enable the operator to operate the machine in a mediated manner and obtain the desired services from the machine, are increasingly being used. Accordingly, in a variety of fields, including but not limited to online counseling centers, online shopping malls, and the like, the user can obtain desired services through an interactive AI agent system that provides natural language conversations in the form of voice and / or text. You can be provided.

The interactive AI agent system builds and utilizes various knowledge base models to understand the user's intent according to the input from the user and to proceed the appropriate dialogue accordingly. In particular, in recent years, beyond the conventional interactive AI agent system that provides only simple scenario-based dialogue service based on fixed scenarios, the interactive AI agent system provides more complex domain service based on free speech type voice input. The demand for it is increasing. Therefore, the interactive AI agent system needs to continuously learn to construct and update various knowledge base models, and the data for learning are recorded as conversation logs by conversations between real people, for example, conversations between real people. Conversation logs on a recording (for example, a recording of a conversation between a customer and a counselor of a counseling center) and the like are typically used.

By the way, the recording of the conversation between the actual people may include a variety of entity information, such as a variety of personal information, such as name, social security number, user ID, telephone number, email, address, and the like. Such information may be information that is required to be de-identified by the requirements of various other laws including the Personal Information Protection Act or by various other needs. In addition, this information usually does not aid in learning for the interactive AI agent system by itself.

Therefore, before learning the interactive AI agent system using the recording, it is necessary to delete or perform the appropriate de-identification processing described above that needs to be de-identified. Such processing needs to be carried out so as not to impair the purpose of learning.

According to one aspect of the present disclosure, a method performed by a computer device is provided. The method of the present disclosure is for automatically building or updating a knowledge base model for an interactive AI agent system, the method comprising: receiving a series of conversation logs associated with each other; Identifying, from each of the received chat logs, entity information determined that de-identification is needed according to predetermined criteria, wherein the entity information includes an entity type and a value; Replacing the value of each identified entity information with each corresponding label-the label being an identifier that identifies the entity type of the corresponding entity information, the entity having the same value of the same type included in the conversation logs The information is replaced with the same label; And building or updating a knowledgebase model for the interactive AI agent system, in accordance with learning using the conversation logs, wherein the value of the identified entity information is replaced with each corresponding label.

According to an embodiment of the present disclosure, the entity information determined to be de-identified includes personal information, and the personal information includes at least one of a name, social security number, date of birth, address, age, telephone number, ID, and email address. It may include.

According to one embodiment of the present disclosure, identifying the entity information determined to be de-identified may include identifying entity information that is determined to be de-identified based on each description format of the personal information. have.

According to one embodiment of the present disclosure, replacing the value of each identified entity information with each corresponding label comprises: replacing entity information with different values of the same type included in the conversation logs with different labels. It may include the step.

According to one embodiment of the present disclosure, receiving the conversation logs may include receiving the conversation logs from a conversation recording between the customer and the agent of the customer center.

According to an embodiment of the present disclosure, the knowledgebase model may include a knowledgebase model for conversation understanding and conversation management.

According to another aspect of the present disclosure, a computer readable recording medium including one or more instructions, wherein the one or more instructions, when executed for a computer, cause the computer to perform any one of the methods described above. A computer readable recording medium is provided.

According to another aspect of the present disclosure, a computer device for automatically building or updating a knowledge base model for an interactive AI agent system, comprising: a conversation log collector configured to collect and store a series of conversation logs associated with each other; And a knowledge base model building / update unit. The knowledge base model building / update unit of the present disclosure receives a series of chat logs from the chat log collecting unit, and from each of the received chat logs, it is determined that de-identification is necessary according to a predetermined criterion-entity information. Includes an entity type and a value, and replaces the value of each identified entity information with each corresponding label, where the label is an identifier that identifies the entity type of the corresponding entity information, The entity information with the same value of the same type included among the components is replaced with the same label, and according to the learning using the conversation logs, the value of the identified entity information is replaced with each corresponding label. It is configured to build or update a knowledge base model for the application.

According to an embodiment of the present disclosure, when learning an interactive AI agent system using, for example, a recording, the legal or non-identifiable information appearing on the recording can be provided without compromising the purpose of learning for the interactive AI agent system. De-identification can be done to meet other needs.

1 is a diagram schematically illustrating a system environment in which an interactive AI agent system may be implemented, according to an embodiment of the present disclosure.

2 is a functional block diagram schematically illustrating the functional configuration of the user terminal 102 of FIG. 1, according to an embodiment of the disclosure.

3 is a functional block diagram schematically illustrating the functional configuration of the interactive AI agent server 106 of FIG. 1, in accordance with an embodiment of the present disclosure.

4 is a functional block diagram schematically illustrating a functional configuration of the conversation / task processing unit 304 of FIG. 3, according to an embodiment of the present disclosure.

FIG. 5 is an exemplary operational flow diagram performed by the knowledgebase model build / update unit 308 of FIG. 3, in accordance with an embodiment of the present disclosure.

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the following, when it is determined that there is a risk of unnecessarily obscuring the subject matter of the present disclosure, a detailed description of already known functions and configurations will be omitted. In addition, it should be understood that what is described below is only related to an embodiment of the present disclosure, but the present disclosure is not limited thereto.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. For example, a component expressed in the singular should be understood as a concept including a plurality of components unless the context clearly indicates the singular. It is to be understood that the term "and / or" as used in this disclosure encompasses all possible combinations by one or more of the listed items. The terms 'comprise' or 'having' as used in the present disclosure are merely intended to designate that there exists a feature, number, step, operation, component, part, or combination thereof described on the present disclosure. It is not intended to exclude the possibility of the presence or the addition of one or more other features or numbers, steps, actions, components, parts or combinations thereof by use.

In the exemplary embodiment of the present disclosure, the module or the unit means a functional part that performs at least one function or operation, and may be implemented in hardware or software or in a combination of hardware and software. Also, a plurality of 'modules' or 'units' may be integrated by at least one software module and implemented by at least one processor, except for 'modules' or 'units', which need to be implemented by specific hardware. have.

In an embodiment of the present disclosure, the interactive AI agent system may be a natural language input (eg, a natural language) input from a user through an interactive interaction with a user through a natural language in the form of voice and / or text. Receive and interpret commands, statements, requests, questions, etc.) from a user who has been determined to determine the user's intent and perform the necessary actions based on that user's intent, i.e. provide appropriate conversational response and / or Or any information processing system capable of providing performance of a task, without being limited to a particular form.

In an embodiment of the present disclosure, the dialogue response provided by the 'interactive AI agent system' may be in visual, auditory and / or tactile form (eg, voice, sound, text, video, image, symbol, emoticon, hyperlink, It may be included in various forms such as, but not limited to, animation, various knots, motion, haptic feedback, and the like. Tasks performed by the interactive AI agent system in embodiments of the present disclosure may, for example, retrieve information, proceed with payment, create a message, create an email, make a phone call, play music, take a picture, search for a user's location, map / Various types of tasks (including, but not limited to, navigation services) may be included.

In an embodiment of the present disclosure, an interactive AI agent system is a chatbot system based on a messenger platform, for example, a chatbot system that exchanges messages with a user on a messenger and provides various information desired by the user or performs a task. It is to be understood that the present disclosure is not limited thereto.

In addition, all terms used in this disclosure, including technical or scientific terms, unless otherwise defined, have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. It is to be understood that the terms used in the dictionary, which are generally used, should be interpreted as having a meaning consistent with the contextual meaning of the related art, and shall not be construed as being excessively limited or extended unless clearly defined otherwise in the present disclosure. .

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically illustrating a system environment 100 in which an interactive AI agent system may be implemented, in accordance with an embodiment of the present disclosure. As shown, the system environment 100 includes a plurality of user terminals 102a-102n, a communication network 104, an interactive AI agent server 106, and an external service server 108.

According to one embodiment of the present disclosure, each of the plurality of user terminals 102a-102n may be any user electronic device having a wired or wireless communication function. Each of the user terminals 102a-102n may be various wired or wireless communication terminals including, for example, a smartphone, a tablet PC, a music player, a smart speaker, a desktop, a laptop, a PDA, a game console, a digital TV, a set-top box, and the like. It should be understood that it is not limited to any particular form. According to one embodiment of the present disclosure, each of the user terminals 102a-102n may communicate with the interactive AI agent server 106, that is, send and receive necessary information, through the communication network 104. According to one embodiment of the present disclosure, each of the user terminals 102a-102n may communicate with the external service server 108, that is, transmit and receive necessary information through the communication network 104. According to one embodiment of the present disclosure, each of the user terminals 102a-102n may receive user input in the form of voice and / or text from the outside, and communicate with the interactive AI agent server 106 via the communication network 104. And / or an operation result corresponding to the above user input (eg, providing a specific conversation response and / or a specific task) obtained through communication with the external service server 108 (and / or processing in the user terminals 102a-102n). May be provided to the user.

According to one embodiment of the present disclosure, each of the user terminals 102a-102n may have a conversational response as a result of an operation corresponding to a user input in a visual, audio, and / or tactile form (eg, voice, sound, text, video, And images, symbols, emoticons, hyperlinks, animations, various notes, motions, haptic feedback, and the like, but are not limited thereto. In an embodiment of the present disclosure, performing a task as an operation corresponding to a user input includes searching for information, proceeding with payment, writing a message, writing an email, making a phone call, playing a music, taking a picture, searching for a user's location, and a map / navigation service. And various types of tasks, including, but not limited to, performing such tasks.

According to one embodiment of the present disclosure, communication network 104 may include any wired or wireless communication network, such as a TCP / IP communication network. According to one embodiment of the present disclosure, the communication network 104 may include, for example, a Wi-Fi network, a LAN network, a WAN network, an Internet network, and the like, but the present disclosure is not limited thereto. According to one embodiment of the present disclosure, the communication network 104 may include, for example, Ethernet, GSM, Enhanced Data GSM Environment (EDGE), CDMA, TDMA, OFDM, Bluetooth, VoIP, Wi-MAX, Wibro, and any other various wired or wireless. It may be implemented using a communication protocol.

According to one embodiment of the present disclosure, the interactive AI agent server 106 may communicate with the user terminals 102a-102n via the communication network 104. According to an embodiment of the present disclosure, the interactive AI agent server 106 transmits and receives necessary information with the user terminals 102a-102n through the communication network 104, and thereby receives them on the user terminals 102a-102n. The operation result corresponding to the input user input, that is, the user intent, may be provided to the user. According to an embodiment of the present disclosure, the interactive AI agent server 106 receives user natural language input in the form of voice and / or text from the user terminals 102a-102n, for example, via the communication network 104, and is prepared in advance. Based on the knowledge base model, the received natural language input can be processed to determine the user's intent. According to an embodiment of the present disclosure, the interactive AI agent server 106 may perform an operation corresponding to the user intent determined above based on a knowledge base model for conversation flow management prepared in advance. According to one embodiment of the present disclosure, the interactive AI agent server 106 obtains from conversation logs (eg, transcripts of a plurality of user and / or system utterance records, etc.) obtained from conversations between real people through various paths. Collected chat logs) and learn the collected chat logs to build and update various knowledge base models for intent determination of user input and / or management of chat flow.

According to one embodiment of the present disclosure, the interactive AI agent server 106 may generate a specific conversation response that corresponds to, for example, a user intent, and send it to the user terminals 102a-102n. According to an embodiment of the present disclosure, the interactive AI agent server 106 generates, based on the user intent determined above, a corresponding conversation response in voice and / or text form, and generates the generated response in a communication network ( Via 104, it may be delivered to the user terminals 102a-102n. According to one embodiment of the present disclosure, the conversation response generated by the interactive AI agent server 106, along with the above-described natural language response in the form of voice and / or text, other visual elements such as images, videos, symbols, emoticons, etc. Or other acoustic elements such as sound, sound, or other tactile elements.

According to one embodiment of the present disclosure, the same type of response on the interactive AI agent server 106, depending on the type of user input received on the user terminals 102a-102n (eg, voice or text input). May be generated (eg, a voice response is generated if a voice input is given and a text response is generated if a text input is given), but the present disclosure is not so limited. In accordance with another embodiment of the present disclosure, it should be appreciated that responses in the form of voice and / or text may be generated and provided regardless of the form of user input.

According to one embodiment of the present disclosure, the interactive AI agent server 106 may communicate with the external service server 108 via the communication network 104, as mentioned above. The external service server 108 may be, for example, a messaging service server, an online consultation center server, an online shopping mall server, an information retrieval server, a map service server, a navigation service server, and the like, but the present disclosure is not limited thereto. According to one embodiment of the present disclosure, a conversation response based on user intent, delivered from the interactive AI agent server 106 to the user terminals 102a-102n, is retrieved and obtained from, for example, the external service server 108. It should be understood that this may include data content.

In this figure, the interactive AI agent server 106 is shown as a separate physical server configured to communicate with the external service server 108 via the communication network 104, but the present disclosure is not limited thereto. According to another embodiment of the present disclosure, it should be understood that the interactive AI agent server 106 may be included as part of various service servers such as an online consultation center server or an online shopping mall server.

2 is a functional block diagram schematically showing the functional configuration of the user terminal 102 shown in FIG. 1, according to an embodiment of the present disclosure. As shown, the user terminal 102 includes a user input receiving module 202, a sensor module 204, a program memory module 206, a processing module 208, a communication module 210, and a response output module ( 212).

According to an embodiment of the present disclosure, the user input receiving module 202 may input various types of inputs from a user, for example, natural language inputs such as voice inputs and / or text inputs (and additionally other types of inputs such as touch inputs). ) Can be received. According to an embodiment of the present disclosure, the user input receiving module 202 may include, for example, a microphone and an audio circuit, and may acquire a user voice input signal through the microphone and convert the obtained signal into audio data. According to an embodiment of the present disclosure, the user input receiving module 202 may include various pointing devices such as a mouse, a joystick, a trackball, a keyboard, a touch panel, a touch screen, a stylus, and the like. The user may acquire text input and / or touch input signals input from the user through these input devices. According to an embodiment of the present disclosure, the user input received by the user input receiving module 202 may be associated with performing a predetermined task, for example, executing a predetermined application or retrieving predetermined information, but the present disclosure is limited thereto. It is not. According to another embodiment of the present disclosure, the user input received by the user input receiving module 202 may require only a simple conversation response regardless of execution of a predetermined application or retrieval of information.

According to one embodiment of the present disclosure, the sensor module 204 includes one or more different types of sensors, through which the status information of the user terminal 102, such as the physical state of the corresponding user terminal 102, The software and / or hardware state or information about the environment state of the user terminal 102 may be obtained. According to an embodiment of the present disclosure, the sensor module 204 may include, for example, an optical sensor, and detect an ambient light state of the corresponding user terminal 102 through the optical sensor. According to an embodiment of the present disclosure, the sensor module 204 may include, for example, a movement sensor, and detect whether the corresponding user terminal 102 is moved through the movement sensor. According to an embodiment of the present disclosure, the sensor module 204 may include, for example, a speed sensor and a GPS sensor, and may detect a position and / or orientation state of the corresponding user terminal 102 through these sensors. According to another embodiment of the present disclosure, it should be appreciated that the sensor module 204 may include other various types of sensors, including temperature sensors, image sensors, pressure sensors, contact sensors, and the like.

According to an embodiment of the present disclosure, the program memory module 206 may be any storage medium in which various programs, for example, various application programs and related data, which may be executed on the user terminal 102, are stored. According to one embodiment of the present disclosure, the program memory module 206 includes, for example, a dialing application, an email application, an instant messaging application, a camera application, a music playback application, a video playback application, an image management application, a map application, a browser application, and the like. Various application programs, including, and data associated with the execution of these programs can be stored. According to an embodiment of the present disclosure, the program memory module 206 may be configured to include various types of volatile or nonvolatile memory such as DRAM, SRAM, DDR RAM, ROM, magnetic disk, optical disk, flash memory, and the like. .

According to one embodiment of the present disclosure, the processing module 208 may communicate with each component module of the user terminal 102 and perform various operations on the user terminal 102. According to one embodiment of the present disclosure, the processing module 208 may drive and execute various application programs on the program memory module 206. According to one embodiment of the present disclosure, the processing module 208 may receive the signals obtained by the user input receiving module 202 and the sensor module 204 and perform appropriate processing on these signals, if necessary. have. According to one embodiment of the present disclosure, the processing module 208 may perform appropriate processing on a signal received from the outside through the communication module 210 if necessary.

According to one embodiment of the present disclosure, the communication module 210 is configured such that the user terminal 102 communicates with the interactive AI agent server 106 and / or the external service server 108 via the communication network 104 of FIG. 1. Enable communication According to an embodiment of the present disclosure, the communication module 212 is configured such that, for example, signals acquired on the user input receiving module 202 and the sensor module 204 may be interactive AI agents through the communication network 104 according to a predetermined protocol. May be sent to server 106 and / or external service server 108. According to one embodiment of the present disclosure, the communication module 210 may, for example, receive various signals, such as voice and / or the like, from the interactive AI agent server 106 and / or the external service server 108 via the communication network 104. Alternatively, a response signal or various control signals including a natural language response in text form may be received, and appropriate processing may be performed according to a predetermined protocol.

According to an embodiment of the present disclosure, the response output module 212 may output a response corresponding to a user input in various forms such as visual, auditory, and / or tactile. According to one embodiment of the present disclosure, the response output module 212 includes various display devices such as a touch screen based on technologies such as LCD, LED, OLED, QLED, etc., and responds to user input through these display devices. Visual responses such as text, symbols, videos, images, hyperlinks, animations, various notes, and the like, to the user. According to one embodiment of the present disclosure, the response output module 212 includes, for example, a speaker or a headset, and provides an audible response, such as a voice and / or acoustic response, to the user via the speaker or the headset, corresponding to a user input. can do. According to an embodiment of the present disclosure, the response output module 212 may include a motion / haptic feedback generator, thereby providing a tactile response, eg, motion / haptic feedback, to a user. In accordance with one embodiment of the present disclosure, it should be appreciated that the response output module 212 can simultaneously provide any two or more combinations of text response, voice response, and motion / haptic feedback corresponding to user input.

3 is a functional block diagram schematically illustrating the functional configuration of the interactive AI agent server 106 of FIG. 1, in accordance with an embodiment of the present disclosure. As shown, the interactive agent server 106 includes a communication module 302, a chat / task processor 304, a chat log collector 306, and a knowledge base model build / update unit 308.

According to one embodiment of the present disclosure, the communication module 302 is configured such that the interactive AI agent server 106 is connected to the user terminal 102 and / or via the communication network 104 according to a predetermined wired or wireless communication protocol. Enable communication with an external service server 108. According to one embodiment of the present disclosure, the communication module 302 may receive voice input and / or text input from the user, transmitted from the user terminal 102, through the communication network 104. According to one embodiment of the present disclosure, the communication module 302, via or via the communication network 104, together with or separately from voice input and / or text input from the user that has been transmitted from the user terminal 102. Status information of the user terminal 102, which has been transmitted from the terminal 102, may be received. According to one embodiment of the present disclosure, the status information may include, for example, various status information related to the corresponding user terminal 102 at the time of voice input and / or text input (eg, physical state of the user terminal 102, user). Software and / or hardware status of the terminal 102, environmental status information around the user terminal 102, and the like. According to one embodiment of the present disclosure, the communication module 302 may further include a conversation response (eg, voice and / or textual natural language) generated by the interactive AI agent server 106 in response to the received user input. And / or control signals, etc.) and / or control signals may be performed as necessary to communicate to the user terminal 102 via the communication network 104.

According to an embodiment of the present disclosure, the conversation / task processing unit 304 receives a user natural language input from the user terminal 102 through the communication module 302 and based on a predetermined knowledge base model prepared in advance. By processing, the user's intent corresponding to the user natural language input may be determined. According to one embodiment of the present disclosure, the conversation / task processing unit 304 may also provide an operation corresponding to the determined user intent, such as proper conversation response and / or task performance.

According to one embodiment of the present disclosure, the conversation log collecting unit 306 may receive and store each conversation log on the conversation collected by any of various methods. According to one embodiment of the present disclosure, the conversation logs received and stored in the conversation log collector 306 may be recording records of conversations between real people, such as recordings of actual conversations between a customer and an agent of a counseling center. . Thus, each of the conversation logs collected and stored on the conversation log collection unit 306 may be a variety of entity information, in particular laws or other such as various personal information, including name, social security number, user ID, telephone number, email, address, etc. It may contain information that requires de-identification for various reasons.

According to an embodiment of the present disclosure, the knowledge base model building / update unit 308 may perform various types of operations for the interactive AI agent server 106 through learning using the respective conversation logs on the above-described conversation log collection unit 306. Knowledgebase models (eg, various knowledgebase models for intent determination or conversation flow management) can be built and / or updated. However, as mentioned above, each conversation log on the conversation log collecting unit 306 may include various entity information that needs to be de-identified. As mentioned above, such entity information usually does not help the learning of the information content itself. However, if you randomly delete the entity information as described above from the conversation logs for this learning, it may be impossible to obtain the intent from the conversation log itself (e.g. in a conversation between A and B While referring to the name or phone number of a person, while referring to the name and phone number of a third person C, the same type of information, such as the name or phone number, is used when different entity information is used multiple times. If you delete them in bulk, the intents of A and B on each conversation log may be compromised or obscured). According to an embodiment of the present disclosure, the knowledge base model building / updater 308 may perform preprocessing on the entity information included in the conversation log to be learned, prior to learning using the conversation logs. This pre-processing by the knowledge base model building / update unit 308 replaces each entity information with low learning possibility on the above-mentioned conversation log and needs to be de-identified, instead of randomly deleting its value with any other identifier. It may be. For example, when the names of different people appear in a series of conversation logs obtained from conversations between people, the knowledge base model building / update unit 308 may use the name value of each name itself (eg, Kim Chul-soo, Lee Young-hee, Hong Gil-dong, etc.). In place of), the preprocessing can be performed by inserting respective identifiers (e.g., labels such as name 1, name 2, name 3, etc.). By this pre-processing, the knowledge base model building / update unit 308 can de-identify the entity information in the conversation log to the extent that the purpose for learning does not harm. According to one embodiment of the present disclosure, the knowledgebase model building / updating unit 308 may learn using preprocessed conversation logs, and perform construction and / or updating of knowledgebase models based on the result. have.

4 is a functional block diagram schematically illustrating a functional configuration of the conversation / task processing unit 304 of FIG. 3, according to an embodiment of the present disclosure. As shown, the conversation / task processing unit 302 may include a speech-to-text (STT) module 402, a natural language understanding (NLU) module 404, and a user database 406. ), Conversation understanding knowledge base model 408, conversation management module 410, conversation flow management knowledge base model 412, conversation generation module 414, and text-to-speech (TTS) module 416 ).

According to an embodiment of the present disclosure, the STT module 402 may receive a voice input among user inputs received through the communication module 302 and convert the received voice input into text data based on pattern matching or the like. have. According to one embodiment of the present disclosure, the STT module 402 may generate a feature vector sequence by extracting features from a user's voice input. According to an embodiment of the present disclosure, the STT module 402 may include a dynamic time warping (DTW) method, a HMM model (Hidden Markov Model), a GMM model (Gaussian-Mixture Mode), a deep neural network model, an n-gram model, and the like. Based on a variety of statistical models of, a text recognition result, such as a sequence of words, can be generated. According to an embodiment of the present disclosure, the STT module 402 may refer to each user characteristic data of the user database 406 described below when converting the received voice input into text data based on pattern matching. .

According to one embodiment of the present disclosure, the NLU module 404 may receive a text input from the communication module 302 or the STT module 402. According to one embodiment of the present disclosure, the text input received at the NLU module 404 is, for example, a user text input or communication module 302 that was received from the user terminal 102 via the communication network 104 at the communication module 302. ) May be a text recognition result generated by the STT module 402 from a user voice input received in the CPC, for example, a sequence of words. According to one embodiment of the present disclosure, the NLU module 404, along with or after receiving the text input, state information associated with the user input, such as state information of the user terminal 102 at the time of the user input. And the like. As described above, the status information may include, for example, various status information related to the corresponding user terminal 102 at the time of user voice input and / or text input (eg, physical state of the user terminal 102 and software). And / or hardware status, environmental status information around the user terminal 102, and the like.

According to one embodiment of the present disclosure, the NLU module 404 may map the received text input to one or more user intents based on the conversation understanding knowledge base model 408 described below. The user intent here may be associated with a series of action (s) that can be understood and performed by the interactive AI agent server 106 in accordance with the user intent. According to one embodiment of the present disclosure, the NLU module 404 may refer to the aforementioned state information in mapping the received text input to one or more user intents. According to one embodiment of the present disclosure, the NLU module 404 may refer to each user characteristic data of the user database 406 described below in mapping the received text input to one or more user intents.

According to an embodiment of the present disclosure, the user database 406 may be a database that stores and manages characteristic data for each user. According to an embodiment of the present disclosure, the user database 406 may, for example, record a previous conversation of the user, pronunciation characteristic information of the user, user vocabulary preferences, the user's location, a setting language, a contact / friend list, etc. It may include various user characteristic information.

According to one embodiment of the present disclosure, as described above, the STT module 402 refers to each user characteristic data of the user database 406, for example, a pronunciation characteristic for each user, when converting a voice input into text data. You can get more accurate text data. According to one embodiment of the present disclosure, the NLU module 404 determines the user intent more accurately by referring to each user characteristic data of the user database 406, such as each user specific feature or context when determining the user intent. can do.

In this figure, a user database 406 for storing and managing characteristic data for each user is illustrated as being disposed in the interactive AI agent server 106, but the present disclosure is not so limited. According to another embodiment of the present disclosure, a user database for storing and managing characteristic data for each user may exist, for example, in the user terminal 102 and is distributed to the user terminal 102 and the interactive AI agent server 106. It should be understood that the arrangement may be made.

According to one embodiment of the present disclosure, the conversation management module 410 may generate a series of operation flows corresponding thereto according to the user intent determined by the NLU module 404. According to one embodiment of the present disclosure, the conversation management module 310, based on the conversation flow management knowledge base model 412, responds to any action, such as any conversation, in response to a user intent received from the NLU module 404. It may determine whether to perform a response and / or perform a task, and generate a detailed operation flow accordingly.

According to one embodiment of the present disclosure, the conversation understanding knowledge base model 408 may include, for example, a predefined ontology model. According to one embodiment of the present disclosure, the ontology model may be represented, for example, in a hierarchical structure between nodes, each node being a "intent" node or a child linked to an "intent" node corresponding to the user's intent. It may be one of the "property" nodes (child "nodes" nodes that are linked directly to "intent" nodes or relinked to "property" nodes of "intent" nodes). According to one embodiment of the present disclosure, an "intent" node and "attribute" nodes that are directly or indirectly linked to the "intent" node may constitute one domain, and the ontology may consist of a collection of such domains. Can be. According to one embodiment of the present disclosure, the conversation understanding knowledge base model 408 is configured to include domains corresponding to all intents each of which, for example, the interactive AI agent system can understand and perform an operation corresponding thereto. Can be. According to one embodiment of the present disclosure, it should be understood that the ontology model may be dynamically changed by adding or deleting nodes, or modifying relationships between nodes.

According to one embodiment of the present disclosure, intent nodes and attribute nodes of each domain in the ontology model may be associated with words and / or phrases related to corresponding user intents or attributes, respectively. According to one embodiment of the present disclosure, the dialogue understanding knowledge base model 408 comprises an ontology model consisting of a hierarchy of nodes and a set of words and / or phrases associated with each node, for example in the form of a lexical dictionary ( The NLU module 404 may determine the user intent based on the ontology model implemented in the form of a lexical dictionary. For example, according to one embodiment of the present disclosure, when the NLU module 404 receives a text input or a sequence of words, the NLU module 404 may determine which words in which domain in the ontology model are associated with each word in the sequence, Based on such a determination, the corresponding domain, ie user intent, can be determined.

According to an embodiment of the present disclosure, the conversation flow management knowledge base model 412 may include a sequential flow distribution model of conversations or actions required to provide a service corresponding to an intent corresponding to a user input. According to one embodiment of the present disclosure, although not specifically illustrated, the conversation flow management knowledge base model 412 may include a library of conversation patterns corresponding to each intent.

According to one embodiment of the present disclosure, the conversation generation module 414 may generate a necessary conversation response based on the conversation / action flow generated by the conversation management module 410. According to an embodiment of the present disclosure, the conversation generating module 414 may, when generating a conversation response, for example, the user characteristic data (eg, the user's previous conversation record, the user's pronunciation characteristic information, the user's pronunciation characteristic information) of the user database 406 described above. Vocabulary preferences, user's location, setting language, contact / friend list, history of each user's previous conversation, etc.).

According to one embodiment of the present disclosure, the TTS module 416 may receive a conversation response generated to be sent to the user terminal 102 by the conversation generation module 414. The conversation response received at the TTS module 418 may be a natural language or a sequence of words in text form. According to an embodiment of the present disclosure, the TTS module 418 may convert the received text form input into a speech form according to various types of algorithms.

In the embodiment of the present disclosure described above with reference to FIGS. 1 to 4, the interactive AI agent system allows the client-server model between the user terminal 102 and the interactive AI agent server 106, in particular the client, to provide only user input / output functions. Although described as being implemented based on a so-called "thin client-server model," which provides and delegates all other functionality of the interactive AI agent system to the server, the present disclosure is not so limited. According to another embodiment of the present disclosure, the interactive AI agent system should be aware that the functions may be implemented by being distributed between the user terminal and the server, or alternatively, may be implemented as a standalone application installed on the user terminal. . In addition, when the interactive AI agent system distributes the functions between the user terminal and the server according to an embodiment of the present disclosure, the distribution of each function of the interactive AI agent system between the client and the server may vary from embodiment to embodiment. It should be appreciated that other implementations may be made. In addition, in the embodiment of the present disclosure described above with reference to FIGS. 1 to 4, for convenience, a specific module has been described as performing certain operations, but the present disclosure is not limited thereto. According to another embodiment of the present disclosure, it should be understood that the operations described as performed by any particular module in the above description may be performed by a separate module from each other.

In step 502, the knowledge base model build / update unit 308 may receive a group of conversation logs collected in any of a variety of ways on the knowledge base model build / update unit 308.

Then, in step 504, entity information that requires preprocessing, included among the received chat logs, may be identified. According to an embodiment of the present disclosure, for example, entity information requiring preprocessing may be any type of information that needs to be de-identified, such as personal information. According to one embodiment of the present disclosure, the personal information that needs to be de-identified may be, for example, information such as name, social security number, date of birth, address, age, telephone number, ID, email address, and the like, but is not limited thereto. According to an embodiment of the present disclosure, such information may be identified based on, for example, a method of notifying each of these pieces of information.

In step 506, each type of entity information identified above, as indicated on the conversation logs, can be determined, and the respective entity information can be replaced using a different label for each type. According to one embodiment of the present disclosure, for example, when a plurality of name information appears on the conversation logs in the corresponding group, instead of each of these name values (eg, “Kim Chul Soo”, “Lee Young Hee”, “Hong Gil Dong”, etc.), Replace with a label of the form "Name 1", "Name 2", ..., "Name n". At this point, it should be noted that the same name value should be given the same label (i.e., if "Name 1" is assigned to any one of the name values shown in a given conversation log, e.g. "Hong Gil-dong", the corresponding conversation on the conversation group) "Name 1" should also be given to the same name value "Hong Gil-dong" that appears during the log and other chat logs).

Then, in step 508, the knowledge base model building / updater 308 utilizes and uses the conversation logs processed in step 506 (i.e., replacement by labeling the identified entity information) for learning. Accordingly, various knowledge base models for the interactive AI agent system can be built and updated.

As will be appreciated by those skilled in the art, the present disclosure is not limited to the examples described herein but may be variously modified, reconfigured and replaced without departing from the scope of the present disclosure. It should be understood that the various techniques described herein may be implemented by hardware or software, or a combination of hardware and software.

A computer program according to an embodiment of the present disclosure may include a storage medium readable by a computer processor or the like, such as an EPROM, an EEPROM, a nonvolatile memory such as a flash memory device, a magnetic disk such as an internal hard disk and a removable disk, a magneto-optical disk, and It may be implemented in a form stored in various types of storage media, including a CDROM disk. In addition, the program code (s) may be implemented in assembly or machine language. All modifications and changes that fall within the true spirit and scope of this disclosure are intended to be covered by the following claims.

Claims

A method performed by a computer device, the method for automatically constructing or updating a knowledge base model for an interactive AI agent system,

Receiving a series of chat logs associated with each other;

Identifying, from each of the received conversation logs, entity information determined according to predetermined criteria that de-identification is necessary, wherein the entity information includes an entity type and a value;

Replacing each value of the identified entity information with each corresponding label, the label being an identifier that identifies an entity type of the corresponding entity information, the same value of the same type included in the conversation logs. Entity information with a replace with the same label; And

Constructing or updating a knowledgebase model for an interactive AI agent system in accordance with learning using the conversation logs, wherein the value of the identified entity information is replaced with each corresponding label.

Including, the method.
The method of claim 1,

The entity information determined to be de-identified includes personal information, and the personal information includes at least one of a name, social security number, date of birth, address, age, telephone number, ID, and email address.
The method of claim 2,

Identifying the entity information determined to be de-identified includes identifying entity information that is determined to be de-identified based on each description format of the personal information.
The method of claim 1,

Replacing the value of each of the identified entity information with each corresponding label comprises replacing entity information having different values of the same type included in the conversation logs with different labels. .
The method of claim 1,

Receiving the conversation logs, receiving the conversation logs from a conversation recording between a customer and an agent of a customer center.
The method of claim 1,

The knowledge base model includes a knowledge base model for conversation understanding and conversation management.
A computer readable recording medium comprising one or more instructions,

The one or more instructions, when executed for a computer, cause the computer to perform the method of any one of claims 1 to 6.
A computer device for automatically building or updating a knowledge base model for an interactive AI agent system.

A conversation log collector for collecting and storing a series of conversation logs associated with each other; And

Including knowledge base model building / update unit,

The knowledge base model building / update unit receives the series of conversation logs from the conversation log collection unit,

Identify, from each of the received conversation logs, entity information determined that de-identification is needed, the entity information comprising an entity type and a value, according to predetermined criteria,

Replacing the value of each said identified entity information with each corresponding label, said label being an identifier for identifying an entity type of corresponding entity information, having the same value of the same type contained in said conversation logs Entity information is replaced with the same label-

Configured to build or update a knowledgebase model for an interactive AI agent system, in accordance with learning with the conversation logs, wherein the value of the identified entity information is replaced with each corresponding label.

Computer devices.