KR20190074508A - Method for crowdsourcing data of chat model for chatbot - Google Patents

Abstract

The present invention relates to a conversation model for a chatbot, and more specifically, to a method for crowdsourcing data of a conversation model for a chatbot, which collects speaking data and induces to a learning situation based on the collected speaking data, such that spoken data are expanded to expand a conversation model. To this end, the method for crowdsourcing data of a conversation model for a chatbot according to the present invention comprises: a step (a) which user data are inputted; a step (b) of processing the user data inputted in the step (a) in a natural language method; a step (c) of determining recognition credibility of a conversation model corresponding to the user data processed in the step (b); a step (d) of providing a question when the recognition credibility determined in the step (c) is below a reference level, and identifying a cause of the low recognition credibility by using response data of a user with respect to the question; and a step (e) of correcting the conversation model based on the cause of the low recognition credibility identified in the step (d).

Description

[0001] The present invention relates to a method for crowd-sourcing data of a conversation model for a chatbot,

More particularly, the present invention relates to a communication model of a chatbot, and more particularly, to a communication model of a chatbot for chatbots in which the ignition data is expanded and the communication model is expanded, Sourcing method.

 People around the world began to use messengers more than social media such as Facebook, Twitter, and the band. Almost all of your service apps (e.g., Peek, PayPal, Uber, Financial Services, etc.) are served in a messenger app by looking at a messenger app (e.g., KakaoTalk or WitChat). As such, the messenger app is already becoming a lifestyle platform.

Chatbots can be thought of as a form of interactive interface, a robot that speaks on behalf of a person on the chat screen, and you may think that the ARS answering system by phone has come into the messenger. In other words, the iPhone's Siri is a voice bot, and Siri comes into the messenger and talks to the user, not his voice.

However, since conventional chatbots recognize and answer predefined keywords by simple matching, when they do not recognize the user's utterance correctly, they can not respond accordingly, . In addition, there was a problem that the data was not enough to cope with various conversation situations.

US 20140122618 A1

Disclosure of Invention Technical Problem [8] Accordingly, the present invention has been made keeping in mind the above problems occurring in the prior art, and an object of the present invention is to provide a communication crowd sourcing The present invention is directed to providing a method for providing a service to a user.

According to another aspect of the present invention, there is provided a chat crowd sourcing method for an interactive model for a chatbot, the method comprising: (a) inputting user data; (b) processing the user data inputted in the step (a) by a natural language technique; (c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b); (d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And (e) modifying the conversation model based on the low reason identified in step (d).

Preferably, when the recognition reliability determined in step (c) is equal to or higher than the reference value, conversation is performed in the current conversation model flow.

Preferably, the reference value is 0.8.

Preferably, the recognition reliability is low because it is possible to respond to the current conversation model but the data is insufficient or there is no conversation model flow that can process the user data on the current conversation model flow .

Preferably, the modification of the dialogue model of the step (e) may be performed when the reason identified in the step (d) is applicable to the current dialogue model but the data is insufficient, , Adds data to the corresponding conversation model flow, and adds a new conversation flow if there is no conversation model flow that can process the user data in the current conversation model flow.

Preferably, the step of adding the new conversation flow further comprises: (e1) providing a question as to how the next conversation model will lead and collecting the user's answer data for the question; And (e2) providing a question as to how to expand after the question provided in the step (e1), and collecting user's answer data for the question.

Preferably, after step (e2), the reliability is confirmed by asking whether the collected answer data is reliable.

Preferably, the process of adding the data is performed by determining what words are the core words using the nouns that are the core of the user data.

Preferably, the flow comprises a dialog node and a dialog edge. The computer-readable medium of claim 1, further comprising: at least one memory for storing computer-executable instructions,

The computer-executable instructions stored in the at least one memory, the method comprising the steps of: (a) inputting user data; (b) processing the user data inputted in the step (a) by a natural language technique; (c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b); (d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And (e) modifying the conversation model based on the low reason identified in step (d) above, is performed for the chatbot.

According to another aspect of the present invention, there is provided a method for sourcing a data model of an interactive model for a chatbot, the method comprising the steps of: (a) inputting user data; (b) processing the user data inputted in the step (a) by a natural language technique; (c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b); (d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And (e) modifying the conversation model based on the low reason identified in step (d).

According to the present invention, an interactive model capable of responding to various utterances of the user can be produced.

1 is a flowchart showing a data crowd sourcing method of an interactive model for a chatbot according to the present invention.
2 is a diagram illustrating an example of a dialogue model structure for explaining a data crowd sourcing method of an interactive model for a chatbot according to the present invention.
FIG. 3 is a diagram for explaining an example of a data crowd sourcing method of a conversation model for a chatbox according to FIG. 1; FIG.
4 is a view showing an example of learning for showing a process according to FIG. 3. FIG.
FIG. 5 is a diagram for explaining another example of a data crowd sourcing method of an interactive model for a chatbot according to FIG. 1; FIG.
FIG. 6 is a view showing an example of learning for showing a process according to FIG. 5; FIG.
FIG. 7 is a diagram illustrating a case where the performance of a classifier is poor in applying a data crowd sourcing method of an interactive model for a chatbot according to the present invention; FIG.
FIG. 8 is a diagram illustrating a case in which an interactive model can not accommodate a data crowd sourcing method of an interactive model for a chatbot according to the present invention; FIG.
FIG. 9 is a block diagram illustrating a data crowd sourcing apparatus of an interactive model for a chatbot according to the present invention; FIG.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary terms, and the inventor should appropriately interpret the concepts of the terms appropriately It should be interpreted in accordance with the meaning and concept consistent with the technical idea of the present invention based on the principle that it can be defined. Therefore, the embodiments described in this specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and do not represent all the technical ideas of the present invention. Therefore, It is to be understood that equivalents and modifications are possible.

FIG. 1 is a flowchart illustrating a data crowd sourcing method of an interactive model for a chatbot according to the present invention. FIG. 2 is a flowchart illustrating an example of a dialogue model structure for explaining a data crowd sourcing method of an interactive model for a chatbot according to the present invention Fig.

2, the dialog model will be described with reference to FIG. 2. Referring to FIG. 2, the dialog model includes a plurality of dialog nodes (arrows) Indicated by solid lines). The conversation node specifies the current state of the conversation, and is a data structure that contains the contents of the chatbot. The node ID at the conversation node indicates the location of the current conversation for the entire conversation model, and the node content is a collection of sentences that can be output by the chatbot in the current conversation situation, such as a replaceable noun portion (Eg You went to {place /}). The conversation edge connects the conversation nodes to the data structure corresponding to the user's input in the conversation, and tells the user what input node should be moved to the input. That is, the edge ID tells the entire conversation model which direction the current node should move. In the present invention, a conversation flow including a conversation node and a conversation edge may be expressed.

In a method for crowd sourcing data of an interactive model according to the present invention, user data is input as shown in FIG. 1 (S100). The user data may be voice data or text data. In the present invention, voice data and text data are collectively referred to as user data.

When the user data is input, the input user data is processed as a natural language by a natural language technique so that the chatbot can understand it (S110).

Then, the recognition reliability of the conversation model corresponding to the user data processed by the natural language technique is determined (S120). When the recognition reliability is '1', it is determined whether the recognition reliability is equal to or greater than the reference value (S130). Here, the reference value means 0.8.

If the recognition reliability is equal to or higher than the reference value, the dialogue is performed in the current dialogue model flow (S140).

On the other hand, if the recognition reliability is less than the reference value, a question is provided to the user, and the reason why the recognition reliability is low is determined using the user's answer data for the provided question (S150). The reason that the recognition reliability is low is that the current dialogue model is applicable but the data is insufficient or there is no dialogue model flow that can process the user data in the current dialogue model flow.

Subsequently, the dialogue model is modified based on the low reason identified in step S150 (S160). Here, the method for modification differs depending on the reason identified in step S150. If the reason is grasped by the current dialogue model, but the data is insufficient, it is determined which one of the conversation flows included in the current dialogue model , Adds data to the corresponding conversation model flow, and adds a new conversation flow if no conversation model flow exists to process user data in the current conversation model flow.

In addition, the process of adding a new conversation flow provides questions about how the next conversation model will lead and collects the user's answer data for the questions provided. And how it can be extended, and collects user response data for this question.

The process of adding data is done by using the nouns that are the core of the user data and identifying the key words using the questions.

In the present invention, the situation in which data crowd sourcing of the conversation model is performed corresponds to two situations. First, there is a case where the performance of the classifier (conversation relation module) for determining the recognition reliability is poor and the sentence is misrecognized. In the case of both cases, the recognition reliability is low in the classifier of the chatbot. Fig. 7 is a diagram showing a case where the classifier performance is not good. Fig. 8 is a diagram Fig.

FIG. 3 is a diagram showing an example of a data crowd sourcing method of a conversation model for a chatbox according to FIG. 1, in which user data is input, the input user data is processed by a natural language processing technique, and a sentence ' The recognition reliability of the inputted sentence is judged. At this time, if the sentence 'went to Rome' has a recognition reliability higher than the reference value for the node _ 1, a conversation flow is developed to the dialog model of 'node _ 2' as shown in FIG. When a sentence such as 'Have eaten pasta', 'Have you seen colosseum' or 'Have bought clothes' has been input to the node_2, it is determined whether the recognition reliability of the inputted sentences is equal to or more than the reference value . At this time, it is determined according to the reference value whether it corresponds to the conversation flow (node_3, node_4, node_5) included in the current model, and the conversation is performed in the corresponding conversation flow. In this case, if the performance of the classifier (the dialogue management module) is not good, the process of receiving the information about the noun which is the core, and the like is inquired once more about the use of the word by the user, . That is, it is possible to use a sentence such as 'Have You Seen the Coliseum' or 'You Have Shopped' for the core colloquial 'Colosseum' and 'Clothes' as in node_4 to node_5 in FIG. It induces the conversation flow so that the flow can be evolved. In this way, when the recognition reliability of the conversation is low, the user is guided to a natural learning situation in which the conversation data is collected instead of outputting the error sentence, so that the chatbot naturally takes over the situation in which the chatbot can not respond to the conversation.

FIG. 4 is a diagram showing an example of learning for showing the process according to FIG. 3. In the case where the recognition reliability can be transferred to the next conversation flow on the conversation edge of the current conversation flow, the recognition reliability is higher than the reference value. As shown in Fig. In this case, the performance of the chatbot can be improved through the process of re-learning including newly acquired data and labels.

FIG. 5 is a diagram illustrating another example of a data crowd sourcing method of an interactive model for a chatbox according to FIG. 1. FIG. 6 illustrates an example of learning for showing a process according to FIG. 5, When the inputted user data is processed by the natural language processing method and the sentence 'went to Rome' is inputted, the recognition reliability of the inputted sentence is judged. At this time, if the sentence 'went to Rome' has a recognition reliability higher than the reference value for the node_1, as shown in FIG. 5, if the dialogue flow of the node_2 is developed or below the reference value, The reason why the recognition reliability is below the reference value and below the reference value as shown in FIG. 5 is that the sentence inputted to the node_1 is " If there is no dialog model that can not be processed, a process of adding a new conversation model (node_N #) is performed. At this time, the process of adding a new conversation model (node_N # Because this is an unacceptable ignition, we can collect data on how the dialogue model can be expanded by asking how the next speech will lead (C) it may also give a chance to think about the fire about the learners did not think through this process, or the conversation situation does not simply provide a dialogue model to promote the language skills of the learners.

In this way, it is difficult to confirm that the data received by the learner is correct when collecting data that extends the dialogue model. In the future, users are asked whether the data of the corresponding node is reliable data, And then added to the conversation model, as shown in FIG.

In addition, if the conversation model is disconnected while expanding the conversation, the conversation model converges to a single point, and thus, the conversation switching topic such as 'but ...' .

9 is a block diagram briefly showing a chatbot for data crowd sourcing of an interactive model according to the present invention.

The chatbot can be implemented as a computer program and executed by a computing device. The chatbot of the present invention implemented as a computer program may be a general computer system.

1, the chatbot 10 may include one or more processors 100, one or more memories 200, one or more input devices 300, one or more output devices 400, one or more communication devices 500 , One or more storage devices 700, and a communication channel 600 for communication therebetween.

The processor 100 is configured to execute instructions within the chatbot 10. The memory 200 is a computer-readable temporary storage medium that does not retain stored data unless power is supplied. The memory 200 is used for temporarily storing, for example, a command for the processor 100 to execute, data necessary for executing the command, data generated by execution of the command, and the like.

The input device 300 is used to receive various types of data from a user. Various types of data may be used depending on the type of data and the computing device 10. For example, a keyboard and a mouse may be used when the computing device 10 is a general computer system, and a keyboard displayed on a touch display may be used if the computing device 10 is a tablet computer. A camera may be used when the input data is a video, and a microphone may be used when the input data is audio.

The output device 400 is used to provide various types of data to the user, and various types of data may be used depending on the type of data and the computing device 10.

The computing device 10 communicates with an external device or network using the communication device 500. Communication using the communication device 500 can be performed by various methods such as wire and wireless.

Storage 700 is a storage medium, such as a hard disk, for long term storage that is computer readable. The computing device 10 stores an operating system 20 that may be executed by the processor 100 and the operating system 20 controls the operation of the computing device 10 and the execution of the computer program. The storage device 700 may store computer programs and / or data that is executed by the computing device 10.

The storage 700 includes a plurality of modules including a computer program executed by the computing device 10 such as a chat management module 710, a natural language processing module 720, a conversation management module 730, a learning management module 750). Also, the storage device 700 is provided with an interactive model database necessary for executing a computer program stored in the plurality of modules.

Some of the modules in the chatbot 10 may be included in the user terminal or some of them may be included in the server (not shown), and the chat management module 710 may be a module for processing the user interface, , A computer program, etc., for the purpose of temporary or permanent access to the physical and virtual medium. The user interface includes physical hardware and logical hardware components and allows the user to manipulate the system and output to indicate the results of what the system is using.

The natural language processing module 720 divides the input text data into morpheme units, which are the smallest units, and analyzes what parts of the morpheme have, and inputs morphemes, verbs, adjectives, And it is used for semantic relationship analysis through parsing process. Parsing is done with specific criteria such as noun phrases, verbal phrases, and adjective phrases, and analyzes how there is a relationship between each chunk. By giving these parsing, you can identify object and modifiers. This information is meaningful in itself, but its use is enhanced by additional analysis of the legend. An analysis is an analysis of the intention of a user utterance. It is a classification of the intention of the user whether it is a question, a request, or a simple emotion expression.

The conversation management module 730 determines whether or not the conversation should be performed or responded to the inputted user data according to the recognition reliability, so that the conversation processing process for the proper answer is performed.

The learning management module 740 improves the performance of the chatbot by re-learning including the acquired data and the label.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. It will be understood that various modifications and changes may be made without departing from the scope of the appended claims.

10: Chambot
20: Operating System
100: Processor
200: memory
300: input device
400: Output device
500: communication device
600: communication channel
700: storage device
710: Chat management module
720: Natural language processing module
730: Dialogue management module
740: Learning Management Module

Claims (11)

As a data crowd sourcing method of a conversation model for chatbots,
(a) inputting user data;
(b) processing the user data inputted in the step (a) by a natural language technique;
(c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b);
(d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And
(e) modifying the conversation model based on the low reason identified in step (d)
A method for crowd sourcing data of a chat model for chatbots. The method according to claim 1,
If the recognition reliability determined in step (c) is equal to or higher than the reference value, conversation is performed in the current conversation model flow
A method for sourcing data crowds of chat models for chatbots featuring. The method according to claim 1,
The reference value means 0.8
A method for sourcing data crowds of chat models for chatbots featuring. The method according to claim 1,
The reason that the recognition reliability is low is that it is possible to respond to the current conversation model but the data is insufficient or the conversation model flow that can process the user data in the current conversation model flow does not exist
A method for sourcing data crowds of chat models for chatbots featuring. The method according to claim 1,
The dialogue model modification of step (e)
If the reason identified in the step (d) is applicable to the current conversation model but the data is insufficient, it is determined which one of the conversation flows included in the current conversation model, And if there is no conversation model flow that can process the user data in the current conversation model flow, adding a new conversation flow
A method for sourcing data crowds of chat models for chatbots featuring. The method of claim 5,
The process of adding the new conversation flow comprises:
(e1) providing a question as to how the next conversation model will lead and collecting user's answer data for the question; And
(e2) providing a question as to how to be expanded after the question provided in the step (e1) and collecting user's answer data for the question
The method comprising the steps of: The method of claim 6,
After the step (e2), the reliability is confirmed by asking whether the collected answer data is reliable
A method for sourcing data crowds of chat models for chatbots featuring. The method of claim 6,
The process of adding the data
The key to user data is by using the nouns to identify the key words
A method for sourcing data crowds of chat models for chatbots featuring. Claim 2
The flow includes a dialog node and a dialog edge
A method for sourcing data crowds of chat models for chatbots featuring. At least one memory for storing computer executable instructions,
The computer-executable instructions stored in the at least one memory, the method of sourcing a data model of a conversation model for a chatbot by the at least one processor,
(a) inputting user data;
(b) processing the user data inputted in the step (a) by a natural language technique;
(c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b);
(d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And
(e) modifying the conversation model based on the low reason identified in step (d)
The apparatus for crowd sourcing data of a dialogue model for a chatbot. A method for sourcing a data model of a conversation model for a chatbot stored in a non-temporary storage medium and executed by a processor,
(a) inputting user data;
(b) processing the user data inputted in the step (a) by a natural language technique;
(c) determining recognition reliability of the conversation model corresponding to the user data processed in the step (b);
(d) providing a question if the recognition reliability determined in step (c) is less than a reference value and determining a reason for low recognition reliability using the user's answer data for the question; And
(e) modifying the conversation model based on the low reason identified in step (d)
The method comprising the steps of:

Images