KR20220085221A - The method of managing chatbot conversation flow using script - Google Patents

Abstract

The present invention relates to a chatbot conversation flow management method.
While the conventional chatbot technology focused on dealing with a fixed pattern of interaction that provides an answer to a user's question, in the present invention, the chatbot's dialog processing software provides various types of interaction between the user and the chatbot expected by the designer. made it possible to process it.
The chatbot's dialog processing software of the present invention processes the user's questions inputted from the user UI through the dialog task manager after classifying them for each user through the message handler. The dialog task manager plays the role of dividing and executing the entire complex dialog scenario into component-level unit dialog tasks. The first user question is executed through the intent classification process in the root dialogue task. Each dialogue task according to the intention classification result is executed, and each unit dialogue operation is performed by the method of executing the script defined in the present invention. The unit conversation task performs interaction with the user according to the method of exchanging messages with the user according to a set script or performing natural language processing, logical operation, branching, etc. Let the conversation flow continue.

Description

How to manage chatbot conversation flow using script {THE METHOD OF MANAGING CHATBOT CONVERSATION FLOW USING SCRIPT}

The present invention relates to a technology for automating some of the tasks performed by humans using a chatbot.

Recently, attempts are being made to automate some of the complex tasks performed by humans through chatbots. A chatbot is a technology that allows users to provide necessary information or services through a familiar conversational interface without the user installing a separate specialized program. These chatbots have gone beyond simple question-and-answering and have begun to replace the relatively simple and regular tasks performed by humans. In response to an abstract question or request from the user, the chatbot can accurately recognize the user's hidden intention through repeated conversations and provide a service that meets the user's intention. Furthermore, it is possible to provide information or services that the user did not intend by looking at the response to the service provided to the user and providing other related services.

However, the conventional chatbot only provides a fixed method of interaction, such as providing an answer to a question. This type of chatbot has been effective in the field of automatic counseling that replaces the role of counselors. However, it is difficult to talk to the user first or to respond to the user's sudden reaction.

Our inventors are studying various conversational methods, such as when a user asks a question about the chatbot's response, shifts the topic to another field, or asks for additional explanation when the chatbot's answer is not satisfactory. To this end, it is necessary to design the chatbot's conversation flow management 'adaptively' according to the conversation topic or the user's reaction.

Among the methods for expressing the conversation flow of a chatbot, a script-based conversation flow expression method such as AIML (Artificial Intelligence Markup Language) or Chat Script has been proposed. It was an effort to describe the flow of conversations with users. However, support for control logic that is possible in general programming languages such as branching or repeating according to the user's reaction was limited.

Prior art literature information

1. Korean Patent Publication No. 1840420

2. Korean Patent Publication No. 1830857

3. Korean Patent Laid-Open Publication No. 10-2019-0066988

The inventors of the present invention have completed the present invention as a result of joint research and discussion to solve the above problems.

An object of the present invention is to propose a software system and a conversation flow processing algorithm that enable a chatbot and a user to adapt and respond to a number of predictable situations in the process of interaction through dialogue.

The gist of the present invention is as follows. Complex dialogue scenarios are broken down into smaller, detailed dialogue tasks. Each subdivided conversation task is executed in a software component having an independent thread. Each detailed conversation task can specify the detailed dialog task to be executed next when the action is completed, and a long flow of conversation can proceed by dynamic connection between each detailed dialog task.

Each detailed dialogue task is described in a script language that includes basic functions provided in general programming languages such as procedural flow, branching, and iteration. Through the configuration that interprets and executes the script that describes the conversation flow, the chatbot is designed to communicate with the user in various situations. Through the flow of detailed conversation tasks defined by the designer of the chatbot service and the execution of scripts that describe each detailed conversation task, the chatbot can actively respond to various situations that occur in conversations with users.

On the other hand, other objects not specified in the present invention will be additionally considered within the range that can be easily inferred from the following detailed description and effects thereof.

In order to achieve the above task, the computer device is a component unit having an independent thread for processing a conversation, including one or more unit conversation tasks pre-designed as a dialog script, and chatbot conversation processing that processes the conversation by linking these unit conversation tasks with each other A method for managing a conversation flow of a chatbot having software installed and initiated by a user terminal inputting a user message to the computer device, the method comprising:

The user terminal inputs a user message to the computer device through the user interface,

The message processor of the computer device divides the user message by user and transmits it to the conversation task manager,

The conversation task manager divides the entire complex conversation scenario into each unit conversation task, but if there is a unit conversation task that is already executed when receiving a user message, the user message is delivered to all the running unit conversation tasks, and the user message is transmitted. If there is no unit conversation operation when receiving, classify the intent of the user message to determine which unit conversation operation corresponds to, and execute the corresponding unit conversation operation;

It is characterized in that the dialog script executor sequentially executes a predefined dialog script in units of commands for each unit dialog task, and designates a subsequent unit dialog task, and dynamically connects each unit dialog task.

In the chatbot conversation flow management method according to a preferred embodiment of the present invention, a separate message sending/receiving module and a conversation flow control module are configured for each unit conversation task, and the message sending/receiving module is configured to send and receive messages to and from a user terminal. It functions as a communication port, and the conversation flow control module functions as a control module port for transferring a conversation flow to a subsequent unit conversation task.

In the chatbot conversation flow management method according to a preferred embodiment of the present invention, the unit conversation task may further include performing intention classification using a predefined command.

In addition, in the chatbot conversation flow management method according to a preferred embodiment of the present invention, the dialogue script is read from the dialogue script storage configured separately for each unit dialogue task.

In addition, it is preferable that the dialog script include Send, Receive, AskListen, Jump, Branch, Exit, Logic, GetEntity, GetIntent, and Http commands.

The chatbot according to the present invention can adaptively respond to a variety of situations predictable by a designer, and a user can receive desired information or services more accurately and in detail by using the chatbot. Rather than a simple one-time Q&A, the user's intentions are more clearly communicated to the chatbot through repeated conversations, and the chatbot can provide the exact service that the user wants.

In addition, the chatbot of the present invention may continue a conversation by a method of additionally providing information determined to be of interest to the user, or may continue a conversation related to a previous conversation. Through this, users can increase their satisfaction in conversations through the chatbot.

On the other hand, even if it is an effect not explicitly mentioned herein, it is added that the effects described in the following specification expected by the technical features of the present invention and their potential effects are treated as described in the specification of the present invention.

1 schematically shows a system configuration showing a chatbot conversation processing structure according to an embodiment of the present invention.
2 schematically shows a functional module structure of a unit conversational task 150 according to an embodiment of the present invention.
3 illustrates a script of the root conversation task 150a.
Fig. 4 shows the linkage between dialogue tasks in a scenario to which the method of the present invention is applied.
5 exemplifies a script for a weather response conversation task in the scenario of FIG. 4 .
6 is a diagram illustrating a conversation between a user and a chatbot using the conversation task of FIGS. 4 and 5 .
※ It is revealed that the accompanying drawings are exemplified as a reference for understanding the technical idea of the present invention, and the scope of the present invention is not limited thereby.

The chatbot dialog processing software of the present invention processes the user's questions inputted from the user UI through the dialog task manager after being divided by user through a message processor and delivered. The dialog task manager plays the role of dividing and executing the entire complex dialog scenario into component-based unit dialog tasks. The first user question is executed through the intent classification process in the root dialogue task. Each dialogue task corresponding to the intention classification result is executed, and each unit dialogue operation is performed by the method of executing the script defined in the present invention. The unit conversation task performs interaction with the user according to the method of exchanging messages with the user according to a set script or performing natural language processing, logical operation, branching, etc. It is structured so that the flow of conversation continues.

Hereinafter, the configuration of the present invention guided by various embodiments of the present invention and effects resulting from the configuration will be described with reference to the drawings. In the description of the present invention, if it is determined that related known functions are obvious to those skilled in the art and may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted.

1 schematically shows a system configuration according to a preferred embodiment of the present invention.

A conversation is made between the user terminal 1 and the computer device 100 . The user terminal 1 may be a phone that delivers a user message by voice, or a computer that delivers a user message by voice or text. The computer device 100 of the present invention is a device composed of one or more hardware and software, and is a chatbot terminal that receives and interprets a user message transmitted by the user terminal 1 to conduct a conversation with the user, and the chatbot conversation processing software is installed has been The method of the present invention is executed by an algorithm determined by this chatbot conversation processing software.

The chatbot conversation processing software of the present invention includes a user interface 110 , a message handler 120 , a conversation task manager 130 , and a conversation flow flower 15 . The conversation flow 15 is composed of one or more unit conversation tasks 150 . In this unit dialog task 150, a dialog script is registered in advance as a component unit having an independent thread for processing a dialog. In addition, the chatbot conversation processing software of the present invention processes each unit conversation task 150 and processes a long conversation between the user and the chatbone while connecting these unit conversation tasks 150 to each other.

As shown in FIG. 1 , a user message input by the user terminal 1 through the user UI 110 is transmitted to the message processor 120 . The message processor 120 serves to classify messages of several users for each user and deliver them to the unit conversation task 150 . The message processor 120 delivers the user message received from the user to the conversation task manager 130 , and serves to transfer the chatbot message generated in the unit conversation task 150 to the user terminal 1 . The dialog task manager 130 manages the execution cycle of several unit dialog tasks 150 existing as software components. In addition, the conversation task manager 130 is responsible for executing the requested conversation task when a conversation request occurs.

If there is no unit conversation task to be executed when a message is received from the user, the root conversation task 150a is executed. 150). The root conversation task 150 is a dialog task that is first executed when a message is received from the user, and classifies the intent of the message received from the user to determine which unit dialog task 150 the received message corresponds to, and the corresponding unit dialog task 150 ) to execute.

A separate message transmission/reception module 151 and a conversation flow control module 152 are configured for each unit conversation task 150 . The message sending/receiving module 151 functions as a communication port for sending and receiving messages to and from the user terminal 1, and the conversation flow control module 152 transfers the conversation flow to a subsequent unit conversation task to be executed after each conversation task is finished. It functions as a control module port to

2 exemplarily shows the internal structure of a unit conversation operation 150 according to an embodiment of the present invention.

A dialog script predefined by the chatbot designer for each unit dialog task 150 is read from the dialog script storage 159 and sequentially executed in the dialog script executor 153 in units of commands.

In addition, when intention classification and entity recognition are required while executing a conversational task, the intention classification result and entity recognition result obtained through the intention classifier 155 and the entity recognizer 156 are communicated using commands defined in the dialog task component, respectively. use it for

Conversation script consists of a total of 10 commands for message transmission/reception, flow control, natural language processing, API call, etc., and can describe interactions with users by combining commands. These commands include Send, Receive, AskListen, Jump, Branch, Exit, Logic, GetEntity, GetIntent, and Http.

The command Send is a command that sends a message to the user terminal, and its arguments include message (variable contents or message to be sent), choices (message to be displayed on the selection button), and callbackMessages (message to be delivered to the chatbot when the selection button is pressed). .

The command Receive is a command to receive a message from a user terminal, and its arguments include key (key to store the received message) and timeout (time to wait (seconds) until message is received).

The command AskListen is a command that sends a message to the user terminal and waits for a user response, and its arguments include key (name of a variable to store the message to be received), message (message to be sent (text)), and choice (message to be displayed on the selection button - semicolon). separated), and callbackMessages (message to be sent to the chatbot when the select button is pressed - separated by semicolons).

Command Jump is a command to move to a specific label, and its arguments include label (label of the statement to be moved) and loopLimit (maximum number of repetitions).

Command Branch is a command that branches to a specific label when the condition variable is the same as the comparison variable (reference), and its arguments are reference (name of the variable to be compared with the condition calculation result), then (referred to when the reference and condition result are the same) The label of the statement) and else (the label of the statement to be moved if the result of the condition is different from the reference).

Command Exit is a command that executes another dialog task after script execution is finished and passes the outputSentence sentence to the input sentence of the next dialog task. variable name that contains the sentence to be passed to the dialog operation).

Command Logic is a command that performs a logical operation according to the logical operation expression and stores the result in the key variable. The result is either true or false. The argument of command Logic includes key (name of variable to store the result of logical operation) and expression (logical expression).

The command GetEntity is a command that extracts an entity from a reference string and stores it in a variable, and its arguments are key (name of the variable to save the entity extraction result), reference (name of the variable where the string to extract entity is stored), entityName (the entity to be extracted). name) is there.

The command GetIntent is a command that performs intent classification on a reference string string and stores it in a variable, and its arguments include key (name of the variable to store the result of intention classification extraction) and reference (name of the variable where the string to extract intention classification is stored in). have.

The command Http is a command that delivers an HTTP request and stores the result in a variable, and its arguments include key (name of a variable to store the request result), URI (request address), and body (body string to be included in the request).

The message received from the user, the processing result of the function, etc. are stored in a variable managed inside the unit conversation job 150, the key value in the function argument is for storing the value in the variable, and the reference value is the value of the stored variable. It is used for reading.

When the conversation task is executed, the message received from the previous conversation task can be obtained by declaring the reference argument of “input sentence”.

For example, a command such as GetIntent (intent classification result, input sentence) classifies the input sentence received from the user and stores the result in a variable value called "intention classification result".

A dialog script is defined as a set of command lines. Each command line is represented by a label, command, and argument.

For example, the root dialog task 150a serves to greet the first user and receive a question from the user, then perform intention classification on the input sentence and execute another dialog task. An example of a script for this dialog operation is shown in FIG. 3 .

In the first line, the command is send and the argument is "Hello, I'm an OO chatbot." When a dialog operation is executed, this message is output to the user.

In the second line, the command is AskListen, the first argument is "question content", and the second argument is "If you have any questions, please ask". The sentence "Ask if you have any questions" is printed on the user UI screen, and when the user inputs a question, the sentence is stored in a variable called "Question content".

In the third line, the command is GetIntent, the first argument is “intent classification result” and the second argument is “question content”. After reading the value in the "question content" variable and performing intention classification, the result is stored in the variable "intention classification result".

The fourth row determines whether the value of the intention classification result is not null and stores it in a variable called "logical operation result", and stores true if the intention classification result is not null, and false if it is null.

In the fifth line, the command is Branch, the first argument is “logical operation result”, the second argument is “intent classification success”, the third argument is “intent classification failure”, and when the logical operation result is true, it is a label of intention classification success. Moves to the row with , and if false, moves to the row labeled 'Intent Classification Failed'.

In the sixth line, the command is Exit, the first argument is "intention classification result", and the second argument is the question content. It finds the next dialog action that corresponds to the content in the "Intent classification result" variable, and executes the next dialog action by passing the value of the "Question content" variable. The current conversation task is ended.

The seventh line is the line executed following the branch of the fifth line, where the command is Send and the argument is "do not understand the question". Show the user the message "I didn't understand the question".

The eighth line moves to the line labeled "question", where the command is Jump, the first argument is "question". That is, the question is asked again.

In the present invention, the entire conversation is configured by connecting the unit conversation tasks 150 to each other. 4 shows an example of a chatbot service scenario introducing weather.

It is assumed that the intention classifier has been trained to give one of the following answers: weather answer, self-introduction, and null according to the content of the question.

When a user asks a question for the first time, as in the process of FIG. 3 , after intention classification in the root dialog task 150a, one of two dialog tasks is executed according to the question content: a weather answer dialog task 150b or self-introduction dialog task 150f do.

5 illustrates the script of the weather answer dialog task 150b. Let's take a look one by one.

In the first line, the command is Send and the argument is "You asked about the weather. Please wait." This message is output to the user when the dialog operation is executed as

In the second line, the command is Http, the first argument is "weather result", and the second argument is "http://api.wisenut.co.kr/weather". Pass an Http fairy and store the result in a variable called "weather result".

In the third line, the command is Send and the argument is "{{weather result}}". In this case, if the argument is {{ }}, the contents stored in the variable are output. Therefore, the contents of the variables stored in the 'weather result' are output to the user.

In the fourth line, the command is AskListen, and the first argument is the name of a variable to store the message to be received as “whether to recommend additional weather information”. The second argument is "Would you like to check more information related to the current weather?" and is a message to send to the user. The third argument is "dressing for the weather, food for the weather, restaurant recommendations, no", which is a message to be displayed on the selection button, separated by semicolons. On the screen, 'dress for the weather', 'food according to the weather', 'recommended restaurant', and 'no' are displayed for the user to select. The fourth parameter is "dressing according to the weather, food suitable for the weather, restaurant recommendation, no", which is a message to be delivered to the chatbot when the selection button is pressed, separated by semicolons. After waiting for the user's response, the user's response is stored in 'Whether to recommend additional weather information'.

In the fifth line, the command is Logic, the first argument is the name of a variable to store the logical operation result as “Logical operation result”, and the second argument is “Whether to recommend additional weather information!=No”, which is a logical operation expression. It is determined whether the value of whether or not to recommend weather additional information is No or not, and stored in a variable called "Logical operation result".

In the sixth line, the instruction is Branch, the first argument is “logical operation result”, the second argument is “recommended”, the third argument is “no recommendation required”, and if the logical operation result is true, there is a label labeled “recommended.” Go to the row, and if false, go to the row labeled No Recommendation.

In the seventh line, the command is Exit, the first argument is "whether to recommend additional weather information", and the second argument is "". It finds the next dialog action that corresponds to the content in the "Whether to recommend additional weather information" variable, and executes the next dialog action by passing the value of the "" variable. The current conversation task is ended.

The eighth line is the line executed after the branch of the sixth line, the command is Send and the argument is "I will end the weather related answer". Shows the message to the user, "I'm going to end weather-related answers."

Similar to the example used in the conversation task of FIG. 5, the external API connection through the Http command was performed in the dress conversation task (150c) suitable for the weather, the food dialog task (150d), and the restaurant recommendation dialog task (150e). It can be implemented for each domain using

FIG. 6 exemplifies a conversation with a user using the dialog tasks of FIG. 4 and the dialog task script defined in FIG. 5 . Of course, a conversation job different from that of FIG. 6 may be processed according to the user's reaction.

For reference, in the chatbot conversation flow management method according to an embodiment of the present invention, the dialog script executor sequentially executes a predefined dialog script in units of commands for each unit conversation task while designating a subsequent unit conversation task. By doing so, it dynamically connects each unit conversational task. This method may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software.

Examples of computer-readable media include hard disks, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks, and ROM, RAM, A hardware device specifically configured to store and execute program instructions, such as flash memory, may be included. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter as well as machine codes such as those generated by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The protection scope of the present invention is not limited to the description and expression of the embodiments explicitly described above. In addition, it is added once again that the protection scope of the present invention cannot be limited due to obvious changes or substitutions in the technical field to which the present invention pertains.

Claims (5)

The computer device is a component unit having an independent thread for processing a conversation, and includes one or more unit conversation tasks pre-designed as a dialog script, and a chatbot conversation processing software that processes the conversation by linking these unit conversation tasks with each other is installed; A chatbot conversation flow management method initiated by a terminal inputting a user message to the computer device, comprising:
The user terminal inputs a user message to the computer device through the user interface,
The message processor of the computer device divides the user message by user and transmits it to the conversation task manager,
The conversation task manager divides the entire complex conversation scenario into each unit conversation task, but if there is a unit conversation task that is already executed when receiving a user message, the user message is delivered to all the running unit conversation tasks, and the user message is transmitted. If there is no unit conversation operation when receiving, classify the intent of the user message to determine which unit conversation operation corresponds to, and execute the corresponding unit conversation operation,
A dialog script executor for each unit dialog task, comprising the step of sequentially executing a predefined dialog script in command units and designating a subsequent unit dialog task, characterized in that each unit dialog task is dynamically connected, How chatbots manage conversation flow.
The method of claim 1,
A separate message sending/receiving module and a dialog flow control module are configured for each unit conversation task, the message sending/receiving module functions as a communication port for sending and receiving messages to a user terminal, and the dialog flow control module is configured as a subsequent unit conversation task A method for managing a conversation flow of a chatbot, which functions as a port of a control module that transmits the conversation flow.
The method of claim 1,
The method for managing a conversation flow of a chatbot further comprising the step of performing intention classification using a command in which the unit conversation task is predefined.
The method of claim 1,
The dialog script is read from a dialog script storage configured separately for each unit dialog task, the chatbot conversation flow management method.
The method of claim 1,
The dialog script is a method for managing a conversation flow of a chatbot that includes Send, Receive, AskListen, Jump, Branch, Exit, Logic, GetEntity, GetIntent, and Http commands.

Images