KR20190125154A - An apparatus for machine learning the psychological counseling data and a method thereof - Google Patents

Abstract

The present invention relates to an apparatus and a method which acquire a learning model by machine learning of psychological counseling data of an expert counselor, and apply a machine learning model to text written by a user to automatically generate psychological counseling text to the user. The automatic conversation method with machine learning of psychological counseling data comprises: a step of receiving text related to psychological analysis in conversation text of a counselor and a user; a step of preprocessing the received text for machine learning to acquire a data set; and a step of using a hierarchical encoder-decoder model with an attention mechanism to perform machine learning on the data set to acquire a machine learning model. The received text includes first text including the conversation text of the counselor and the user and second text which is response text of an expert counselor immediately after the first text. The automatic conversation apparatus performs machine learning on a correlation of the first and the second text.

Description

An apparatus for machine learning the psychological counseling data and a method

The present disclosure relates to a method and apparatus for acquiring a learning model by machine learning psychological counseling data of a professional counselor. The present invention also relates to a method and apparatus for automatically generating a psychological counseling text to a user by applying a text written by the user to a learning model.

The mental health care industry is growing rapidly, and there is an increasing demand for automatic communication devices that automatically provide psychological counseling on behalf of professional counselors. In order for the user and the automatic conversation device to talk naturally, the automatic conversation device needs to grasp the user's intention and present the user with a response to a wide range of topics. In addition, the automatic conversation apparatus needs to generate various responses to similar utterances of the user to increase the immersion of the user. In addition, when the automatic conversation device is a machine learning, it is necessary to learn the contents related to psychological counseling, so as to generate a response that does not deviate from the subject of psychological counseling when talking with the user.

Artificial Intelligence (AI) system is a computer system that implements human-level intelligence, and unlike conventional rule-based smart systems, machines learn, judge, and become smart. As the AI system is used, the recognition rate is improved and the user's taste can be understood more accurately. The existing Rule-based smart system is gradually replaced by the deep learning-based AI system.

AI technology is composed of elementary technologies that utilize machine learning (deep learning) and machine learning. Machine learning is an algorithm technology that classifies / learns characteristics of input data by itself, and element technology is a technology that simulates the functions of human brain cognition and judgment by using machine learning algorithms such as deep learning. It consists of technical fields such as prediction and expression of knowledge.

The various fields in which artificial intelligence technology is applied are as follows. Linguistic understanding is a technology for recognizing and applying / processing human language / characters and includes natural language processing, machine translation, dialogue system, question and answer, speech recognition / synthesis, and the like. Inference Prediction is a technique for judging, logically inferring, and predicting information. It includes knowledge / probability-based inference, optimization prediction, preference-based planning, and recommendation. Knowledge expression is a technology that automatically processes human experience information into knowledge data, and includes knowledge construction (data generation / classification) and knowledge management (data utilization).

The present disclosure relates to a method and apparatus for acquiring a learning model by machine learning dialogues of a user and a professional counselor. In addition, the present invention relates to an automatic conversation apparatus and method for generating a response to a user's dialogue based on a learning model, thereby naturally communicating with the user.

An automatic conversation apparatus according to an embodiment of the present disclosure may include a processor and a memory in which at least one program is stored. The memory includes, when the at least one program is executed in the processor, the processor receiving texts related to psychoanalysis among conversation texts of the counselor and the user, preprocessing the received texts for machine learning to obtain a data set; and Stores instructions to execute the step of acquiring the machine learning model by machine learning the data set using a hierarchical encoder-decoder with attention mechanism. The received text is stored by the counselor. And a first text including a conversation text of a user and a second text which is a response text of a counselor immediately after the first text, and obtaining the machine learning model comprises: a correlation between the first text and the second text; It is characterized in that the step of obtaining a machine learning model.

Instructions for causing the processor of the automatic conversation apparatus to execute the step of acquiring the machine learning model according to an embodiment of the present disclosure may include: machine learning about the correlation between the first text, the activity log of the user, and the psychological profile and the second text. Obtaining a model, wherein the user's activity log includes at least one of view posts, comments, empathy, and bookmarks, and the psychological profile includes at least one of the user's gender, age, propensity profile, and personality test results Characterized in that.

Instructions for causing the processor of the automatic conversation apparatus to execute a step of acquiring a data set may include: dividing the received text into a first text and a second text; And dividing the first text and the second text into at least one unit of phoneme, syllable, morpheme, word, word, or sentence.

Instructions for causing the processor of the automatic communication apparatus to execute the step of acquiring the machine learning model according to an embodiment of the present disclosure, by performing the machine learning the correlation between the upper layer unit of the first text and the second text, Obtaining weights for upper layer units of the first text, performing machine learning on correlation between the lower layer units of the first text and the second text, and obtaining weights for the lower layer units of the first text; And obtaining a weight for the modified lower layer unit based on the weight for the upper layer unit and the weight for the lower layer unit. The hierarchical unit is characterized by being a higher hierarchical unit from a lower hierarchical unit toward phonemes, syllables, morphemes, words, words, and sentences.

The memory of the automatic conversation apparatus according to an embodiment of the present disclosure may include receiving, when the at least one program is executed in the processor, the processor receiving first text related to psychoanalysis among conversation texts of the counselor and the user, and receiving the first text. Acquiring a data set by preprocessing the text, and using the hierarchical encoder-decoder with attention mechanism to obtain a data set from the first text received based on a previously learned machine learning model. Store instructions that cause the second text to be automatically acquired and the second text to be output.

Instructions for causing the processor of the automated communication device to automatically acquire the second text according to an embodiment of the present disclosure may include the first text received based on a pre-learned machine learning model, an activity log of a user, and Automatically obtaining a second text from the psychological profile. The activity log of the user may include at least one of a post reading, a comment, empathy, and a bookmark, and the psychological profile may include at least one of a user's gender, age, propensity profile, and personality test results.

Instructions for causing the processor of the automatic conversation apparatus to execute the step of acquiring the data set may include separating the first text into at least one unit of phoneme, syllable, morpheme, word, word, or sentence. Steps.

Instructions for executing the step of automatically obtaining the second text by the processor of the automatic communication apparatus according to an embodiment of the present disclosure, weights for the upper layer unit included in the previously learned machine learning model, the lower layer unit And automatically obtaining the second text based on the weight and the first text, wherein the hierarchical unit is a higher hierarchical unit from a lower hierarchical unit toward phonemes, syllables, morphemes, words, words, and sentences. .

Instructions for causing the processor of the automatic communication device to automatically acquire the second text according to an embodiment of the present disclosure may include: embedding the first text based on a previously learned machine learning model to embed the first text; Obtaining cosine similarity between the embedded first text and the embedding of the plurality of candidates included in the pre-learned machine learning model, and the most similar to the first text among the plurality of candidates based on the calculated similarity. Selecting a second text corresponding to the candidate.

The memory of the automatic conversation apparatus according to an embodiment of the present disclosure stores instructions that, when at least one program is executed in the processor, cause the processor to execute the step of outputting a Polbeck response when the cosine similarity is less than a threshold. .

According to an embodiment of the present disclosure, an automatic conversation method includes receiving texts related to psychoanalysis among conversation texts of a counselor and a user, preprocessing the received texts for machine learning, and obtaining a data set, and a attention mechanism. Acquiring a machine learning model by machine learning the data set using a hierarchical encoder-decoder with attention mechanism, wherein the received text includes conversation text of the counselor and the user. Comprising the first text and the second text which is the response text of the counselor immediately after the first text, and obtaining the machine learning model, obtaining a machine learning model for the correlation between the first text and the second text It is characterized by that.

According to an embodiment of the present disclosure, an automatic conversation method may include receiving a first text related to psychoanalysis among conversation texts of a counselor and a user, obtaining a data set by preprocessing the received first text, and an attention mechanism. Automatically obtaining a second text from the received first text based on a previously learned machine learning model using a hierarchical encoder-decoder with attention mechanism, and outputting the second text It includes a step.

In addition, the program for implementing the mental state prediction method as described above can be recorded in a computer-readable recording medium.

1 is a block diagram of a processor 100 according to an embodiment of the present disclosure.
2 is a block diagram of the data learning unit 110 according to an embodiment of the present disclosure.
3 is a block diagram of the data recognition unit 120 according to an embodiment of the present disclosure.
4 is a diagram illustrating an example in which the device 400 and the server 420 learn and recognize data by interworking with each other according to an embodiment of the present disclosure.
5 is a flowchart illustrating a method of obtaining a learning model according to an embodiment of the present disclosure.
6 illustrates a first text and a second text according to an embodiment of the present disclosure.
7 is a view illustrating a learning unit of an automatic conversation apparatus according to an embodiment of the present disclosure.
8 is a flowchart illustrating a method of generating a response according to a dialogue of a user using a machine learning model according to an embodiment of the present disclosure.
9 is a block diagram of a recognition unit of an automatic conversation apparatus according to an embodiment of the present disclosure.
10 is a view showing an operation screen of the automatic conversation apparatus according to one embodiment of the present disclosure.

Advantages and features of the disclosed embodiments, and methods of achieving them will be apparent with reference to the embodiments described below in conjunction with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below, but may be implemented in various forms, and the present embodiments are merely provided to make the present disclosure complete, and those of ordinary skill in the art to which the present disclosure belongs. It is merely provided to fully inform the scope of the invention.

Terms used herein will be briefly described, and the disclosed embodiments will be described in detail.

The terminology used herein has been selected among general terms that are currently widely used while considering the functions of the present disclosure, but may vary according to the intention or precedent of a person skilled in the relevant field, the emergence of a new technology, and the like. In addition, in certain cases, there is also a term arbitrarily selected by the applicant, in which case the meaning will be described in detail in the description of the invention. Therefore, the terms used in the present disclosure should be defined based on the meanings of the terms and the contents throughout the present disclosure, rather than simply the names of the terms.

A singular expression in this specification includes a plural expression unless the context clearly indicates that it is singular. Also, the plural expressions include the singular expressions unless the context clearly indicates the plural.

When any part of the specification is to "include" any component, this means that it may further include other components, except to exclude other components unless otherwise stated.

Also, as used herein, the term "part" means a software or hardware component, and "part" plays certain roles. However, "part" is not meant to be limited to software or hardware. The “unit” may be configured to be in an addressable storage medium and may be configured to play one or more processors. Thus, as an example, a "part" refers to components such as software components, object-oriented software components, class components, and task components, processes, functions, properties, procedures, Subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables. The functionality provided within the components and "parts" may be combined into a smaller number of components and "parts" or further separated into additional components and "parts".

According to an embodiment of the present disclosure, the “unit” may be implemented with a processor and a memory. The term “processor” should be interpreted broadly to include general purpose processors, central processing units (CPUs), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some circumstances, a “processor” may refer to an application specific semiconductor (ASIC), a programmable logic device (PLD), a field programmable gate array (FPGA), or the like. The term "processor" refers to a combination of processing devices such as, for example, a combination of a DSP and a microprocessor, a combination of a plurality of microprocessors, a combination of one or more microprocessors in conjunction with a DSP core, or a combination of any other such configuration. May be referred to.

The term "memory" should be interpreted broadly to include any electronic component capable of storing electronic information. The term memory refers to random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), programmable read-only memory (PROM), erase-programmable read-only memory (EPROM), electrical May also refer to various types of processor-readable media, such as erasable PROM (EEPROM), flash memory, magnetic or optical data storage, registers, and the like. If the processor can read information from and / or write information to the memory, the memory is said to be in electronic communication with the processor. The memory integrated in the processor is in electronic communication with the processor.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the embodiments. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present disclosure.

1 is a block diagram of a processor 100 according to an embodiment of the present disclosure.

Referring to FIG. 1, the processor 1300 may include a data learner 110 and a data recognizer 120.

The data learner 110 may learn a criterion for generating a response to the dialogue of the user. A conversation (conversation text) is a voice or text that represents the thought of a user or a professional counselor. A conversation is a word spoken by a party from the beginning of the talk after the end of the talk to the end of the talk. The conversation may be one or more sentences. The conversation may be a single word.

The data learner 110 may learn a criterion about what data is used to generate a response to the user's dialogue, and how to determine the mental state using the data. The data learner 110 acquires data to be used for learning and applies the acquired data to a data learning model to be described later, thereby learning a criterion for generating a response to the user's dialogue.

The data recognizer 120 may generate a response to the user's conversation based on the data. The data recognizer 120 may generate a response to the user's dialogue from the predetermined data by using the learned data learning model. The data recognizing unit 120 acquires predetermined data according to a preset criterion by learning, and uses a data learning model using the acquired data as an input value, thereby responding to a predetermined user's dialogue based on the predetermined data. Can be generated. In addition, the result value output by the data learning model using the acquired data as an input value may be used to update the data learning model.

At least one of the data learner 110 and the data recognizer 120 may be manufactured in the form of at least one hardware chip and mounted on the electronic device. For example, at least one of the data learner 110 and the data recognizer 120 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or a conventional general purpose processor (eg, a CPU). Alternatively, the electronic device may be manufactured as a part of an application processor or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In addition, the data learner 110 and the data recognizer 120 may be mounted in separate electronic devices, respectively. For example, one of the data learner 110 and the data recognizer 120 may be included in the electronic device, and the other may be included in the server. In addition, the data learner 110 and the data recognizer 120 may provide the model information constructed by the data learner 110 to the data recognizer 120 through a wired or wireless connection. The data inputted to 120 may be provided to the data learner 110 as additional learning data.

Meanwhile, at least one of the data learner 110 and the data recognizer 120 may be implemented as a software module. When at least one of the data learning unit 110 and the data recognizing unit 120 is implemented as a software module (or a program module including instructions), the software module may be a non-transitory computer readable non-transitory computer. It may be stored in a non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

2 is a block diagram of the data learning unit 110 according to an embodiment of the present disclosure.

Referring to FIG. 2, the data learner 110 may include a data acquirer 210, a preprocessor 220, a training data selector 230, a model learner 240, and a model evaluator ( 250).

The data acquirer 210 may acquire data necessary for generating a response to a user's dialogue. The data acquirer 210 may acquire data necessary for learning to generate a response to a user's dialogue. The data acquirer 210 may receive content of a conversation between a user and a professional counselor.

A user may talk to a professional counselor by text using an electronic terminal such as a PC or a smartphone. In addition, the user may talk to a professional counselor by voice using a voice communication terminal such as a smartphone or a telephone. The data acquirer 210 may receive a conversation content in text. In addition, the data acquisition unit 210 may receive the conversation contents that are spoken and convert them into text. Since a lot of data is required for learning, the data acquirer 210 may receive conversation contents of a plurality of users and a professional counselor.

The preprocessor 220 may preprocess the acquired data so that the obtained data may be used for machine learning to generate a response to the user's dialogue. The preprocessor 320 may process the acquired data into a preset format so that the model learner 240, which will be described later, may use the acquired data to generate a response to the user's dialogue. For example, the preprocessor 320 may divide the conversation contents of the user and the professional counselor into a first text and a second text. The second text is the last text containing contents related to the psychoanalysis of the expert counselor. The first texts are the dialogues of the expert counselor and the user that appeared just before the second text appeared.

The training data selector 230 may select data necessary for learning from the preprocessed data. The selected data may be provided to the model learner 240. The training data selector 230 may select data required for learning from among the preprocessed data according to a preset criterion for generating a response to the user's dialogue. For example, a conversation of a professional counselor may be selected. In addition, a predetermined number of conversations of a professional counselor or a user immediately before a conversation of a professional counselor may be selected. In addition, the training data selector 230 may select data according to preset criteria by learning by the model learner 240, which will be described later.

The model learner 240 may learn a criterion about how to generate a response to the dialogue of the user based on the training data. In addition, the model learner 240 may learn a criterion about what training data should be used to generate a response to the user's dialogue.

In addition, the model learner 240 may learn by using the data learning model used to generate a response to the user's dialogue as the training data. In this case, the data learning model may be a previously built model. For example, the data learning model may be a model built in advance based on basic training data (eg, a sample image).

The data learning model may be constructed in consideration of the application field of the learning model, the purpose of learning, or the computer performance of the device. The data learning model may be, for example, a model based on a neural network. For example, models such as Deep Neural Network (DNN), Recurrent Neural Network (RNN), Long Short-Term Memory models (LSTM), Bidirectional Recurrent Deep Neural Network (BRDNN), and Convolutional Neural Networks (CNN) can be used as data learning models. But it is not limited thereto.

According to various embodiments of the present disclosure, when there are a plurality of pre-built data learning models, the model learning unit 240 may determine a data learning model having a high correlation between the input learning data and the basic learning data as a data learning model to be trained. have. In this case, the basic training data may be previously classified by data type, and the data learning model may be pre-built for each data type. For example, the basic training data is classified based on various criteria such as the region where the training data is generated, the time at which the training data is generated, the size of the training data, the genre of the training data, the creator of the training data, and the types of objects in the training data. It may be.

In addition, the model learner 240 may train the data learning model using, for example, a learning algorithm including an error back-propagation method or a gradient descent method.

In addition, the model learner 240 may learn the data learning model through, for example, supervised learning using the learning data as an input value. In addition, the model learning unit 240 learns data through unsupervised learning that finds a criterion for situation determination by, for example, self-learning a type of data necessary for situation determination without any guidance. You can train the model. In addition, the model learner 240 may learn the data learning model through, for example, reinforcement learning using feedback on whether the result of the situation determination according to the learning is correct.

In addition, when the data learning model is learned, the model learning unit 240 may store the learned data learning model. In this case, the model learner 240 may store the learned data learning model in a memory of the electronic device including the data recognizer 120. Alternatively, the model learner 240 may store the learned data learning model in a memory of a server connected to the electronic device through a wired or wireless network.

In this case, the memory in which the learned data learning model is stored may store, for example, commands or data related to at least one other element of the electronic device. The memory may also store software and / or programs. The program may include, for example, a kernel, middleware, an application programming interface (API) and / or an application program (or “application”), and the like.

The model evaluator 250 may input the evaluation data into the data learning model, and allow the model learner 240 to relearn when the recognition result output from the evaluation data does not satisfy a predetermined criterion. In this case, the evaluation data may be preset data for evaluating the data learning model.

For example, the model evaluator 250 does not satisfy a predetermined criterion when the number or ratio of the evaluation data that is not accurate among the recognition results of the learned data learning model for the evaluation data exceeds a preset threshold. It can be evaluated as not. For example, when a predetermined criterion is defined as a ratio of 2%, when the trained data learning model outputs an incorrect recognition result for more than 20 evaluation data out of a total of 1000 evaluation data, the model evaluator 250 learns. The data learning model can be evaluated as not suitable.

On the other hand, when there are a plurality of trained data learning models, the model evaluator 250 evaluates whether each of the learned video learning models satisfies a predetermined criterion and uses the model satisfying the predetermined criterion as the final data learning model. You can decide. In this case, when there are a plurality of models satisfying a predetermined criterion, the model evaluator 250 may determine any one or a predetermined number of models which are preset in the order of the highest evaluation score as the final data learning model.

At least one of the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 in the data learner 110 may be at least one. May be manufactured in the form of a hardware chip and mounted on an electronic device. For example, at least one of the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 may be artificial intelligence (AI). It may be manufactured in the form of a dedicated hardware chip, or may be manufactured as a part of an existing general purpose processor (eg, a CPU or an application processor) or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In addition, the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 may be mounted on one electronic device or may be separate. Each of the electronic devices may be mounted on the electronic device. For example, some of the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 are included in the electronic device, and the rest of the data is included in the electronic device. Can be included on the server.

In addition, at least one of the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 may be implemented as a software module. A program in which at least one of the data acquirer 210, the preprocessor 220, the training data selector 230, the model learner 240, and the model evaluator 250 includes a software module (or instruction). Module may be stored on a computer readable non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

3 is a block diagram of the data recognition unit 120 according to an embodiment of the present disclosure.

Referring to FIG. 3, the data recognizer 120 according to an exemplary embodiment may include a data acquirer 310, a preprocessor 320, a recognition data selector 330, a recognition result provider 340, and a model updater. And may include 350.

The data acquirer 310 may acquire data necessary for generating a response to the user's conversation, and the preprocessor 320 may use the acquired data to generate a response to the user's conversation. The acquired data can be preprocessed. The preprocessor 320 may process the acquired data into a preset format so that the recognition result providing unit 340, which will be described later, may use the acquired data to generate a response to the user's dialogue.

The recognition data selector 330 may select data required for generating a response to the dialogue of the user from the preprocessed data. The selected data may be provided to the recognition result provider 340. The recognition data selector 330 may select some or all of the preprocessed data according to a preset criterion for generating a response to the user's dialogue. In addition, the recognition data selector 330 may select data according to a predetermined criterion by learning by the model learner 240.

The recognition result providing unit 340 may generate a response to the user's dialogue by applying the selected data to the data learning model. The recognition result providing unit 340 may provide a recognition result according to the recognition purpose of the data. The recognition result providing unit 340 may apply the selected data to the data learning model by using the data selected by the recognition data selecting unit 330 as an input value. In addition, the recognition result may be determined by the data learning model.

The model updater 350 may cause the data learning model to be updated based on the evaluation of the recognition result provided by the recognition result provider 340. For example, the model updater 350 may provide the model learning unit 240 with the recognition result provided by the recognition result providing unit 340 so that the model training unit 240 updates the data learning model. have.

Meanwhile, at least one of the data acquiring unit 310, the preprocessor 320, the recognition data selecting unit 330, the recognition result providing unit 340, and the model updating unit 350 in the data recognizing unit 120 may be at least. It may be manufactured in the form of one hardware chip and mounted on an electronic device. For example, at least one of the data acquirer 310, the preprocessor 320, the recognition data selector 330, the recognition result provider 340, and the model updater 350 may be artificial intelligence (AI). ) May be manufactured in the form of a dedicated hardware chip, or may be manufactured as a part of an existing general purpose processor (eg, a CPU or an application processor) or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In addition, the data acquisition unit 310, the preprocessor 320, the recognition data selector 330, the recognition result providing unit 340, and the model updater 350 may be mounted on one electronic device or may be separate. May be mounted on the electronic devices. For example, some of the data acquirer 310, the preprocessor 320, the recognition data selector 330, the recognition result provider 340, and the model updater 350 may be included in the electronic device, and some of the others may be included in the electronic device. May be included in the server.

In addition, at least one of the data acquirer 310, the preprocessor 320, the recognition data selector 330, the recognition result provider 340, and the model updater 350 may be implemented as a software module. At least one of the data acquirer 310, the preprocessor 320, the recognition data selector 330, the recognition result provider 340, and the model updater 350 may include a software module (or instructions). If implemented as a program module, the software module may be stored in a computer readable non-transitory computer readable media. In this case, at least one software module may be provided by an operating system (OS) or by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and others may be provided by a predetermined application.

4 is a diagram illustrating an example in which the device 400 and the server 420 learn and recognize data by interworking with each other according to an embodiment of the present disclosure.

Referring to FIG. 4, the server 420 may learn a criterion for generating a response to the user's conversation, and the device 400 may respond to the user's conversation based on the learning result by the server 420. Can be generated.

In this case, the model learner 430 of the server 420 may perform the function of the data learner 110 shown in FIG. 2. The model learner 430 of the server 420 may learn a criterion about what data to use to generate a response to the user's conversation and how to generate a response to the user's conversation using the data. . The model learner 430 acquires data to be used for learning, and applies the acquired data to a data learning model to be described later, thereby learning a criterion for generating a response to the user's dialogue.

In addition, the recognition result providing unit 404 of the device 400 may apply the data selected by the recognition data selection unit 403 to the data learning model generated by the server 420 to generate a response to the user's dialogue. Can be. For example, the recognition result providing unit 404 transmits the data selected by the recognition data selection unit 403 to the server 420, and the server 420 learns the data selected by the recognition data selection unit 403. You can apply it to the model and request to generate a response to the user's dialog. In addition, the recognition result providing unit 404 may receive information about the response to the user's dialogue generated by the server 420 from the server 420.

Alternatively, the recognition result providing unit 404 of the device 400 may receive a learning model generated by the server 420 from the server 420, and generate a response to a user's dialogue using the received learning model. Can be. In this case, the recognition result providing unit 404 of the device 400 may apply the data selected by the recognition data selection unit 403 to the data learning model received from the server 420 to generate a response to the user's dialogue. Can be.

4 illustrates a configuration in which the device 400 includes a recognizer 410 and a server 420 includes a learner 430, but the present invention is not limited thereto. The device 400 may be implemented as a server different from the server 420. The user may receive only recognition result information from the device 400.

5 is a flowchart illustrating a method of obtaining a learning model according to an embodiment of the present disclosure.

The automatic communication device may include a processor and a memory in which at least one program is stored. The memory of the automatic conversation apparatus may store at least one program that may be executed in the processor.

The automatic conversation apparatus may perform a step 510 of receiving texts related to psychoanalysis among conversation texts of a professional counselor and a user. The automatic conversation apparatus may perform step 520 of acquiring a data set by preprocessing the received text for machine learning. The automatic conversation apparatus may perform step 530 of acquiring the machine learning model by machine learning the data set using a hierarchical encoder-decoder with attention mechanism. The received text may include a first text including a conversation (conversation text) of the expert counselor and the user, and a second text that is a response text of the expert counselor immediately after the first text. Acquiring the machine learning model (530) may be a step of obtaining a machine learning model for the correlation between the first text and the second text.

The data acquisition unit 210 included in the automatic conversation apparatus may perform a step 510 of receiving texts related to psychoanalysis among conversation texts of a professional counselor and a user. The automatic conversation apparatus may provide the user with the subjective and multiple choice question set Q based on the counseling theory in order to receive the text of the dialogue between the expert counselor and the user. The automatic conversation device may receive a response (R) to the question set from the user. The automatic conversation device may receive a typed response (F) for the user's response (R) from a professional counselor. The professional counselor may provide the user with various responses F according to the consultation manual.

The automatic conversation device may provide a multiple-choice question set Q to a user, receive a response R to the question set from the user, and receive a typed response F from a professional counselor. The question set (Q), the answer (R), and the answer (F) are structured, so that the learning unit of the automatic dialog can perform machine learning quickly and accurately.

In addition, since the automatic dialogue device first presents the question set Q, the automatic dialogue device may limit the subject of the data set for generating the machine learning model to psychological counseling. The recognition unit of the automatic conversation apparatus may generate a response text to the user's conversation text by applying the learned machine learning model to the user's conversation text. Since the data set is typed, the response text of the recognition part of the automatic conversation device may not cause negative reaction to the user, and may deal with the subject of psychological counseling.

The automatic conversation device allows users and professional counselors to receive a log of real-time chat consultation. The automatic dialogue device can learn the real time dialogues of the user and the professional counselors, and acquire the learning model. The automatic conversation device can respond to the free dialogue of the user based on the learning model.

The preprocessing unit 220 included in the automatic conversation apparatus may perform a step 520 of obtaining a data set by preprocessing the received text for machine learning. The automatic conversation apparatus may perform the step of dividing the received text into a first text and a second text. The second text may be the last conversation of the text received in response to a professional counselor. The first text may be at least one conversations before the second text.

The automatic conversation apparatus may perform the step of separating the first text and the second text into at least one unit of phoneme, syllable, morpheme, word, word, or sentence. The automatic conversation apparatus may separate the first text and the second text into at least one unit of a phoneme, syllable, morpheme, word, word, or sentence, and insert a symbol representing a start point and an end point of each unit. For example, the automatic conversation apparatus may divide the first text and the second text into morphemes and insert a symbol representing a start at a start point of one morpheme. You can also insert an end sign at the end of the morpheme.

The symbol for start and the symbol for end can be different. However, the present invention is not limited thereto, and the sign indicating the start and the sign indicating the end may be the same. This is because a symbol indicating the end of one unit may be a symbol indicating the start of another unit.

The automatic conversation apparatus may use different symbols to distinguish different units. For example, a symbol that separates words may be different from a symbol that separates sentences. The automatic conversation apparatus may perform learning using at least one unit of phoneme, syllable, morpheme, word, word, or sentence based on the type of symbol.

The automatic conversation apparatus may add annotation characters to the phonemes, syllables, morphemes, or words in order to distinguish phonemes, syllables, morphemes, or words having the same sounds but having different meanings. For example, "a" may be an investigation or "a" of "go". The automatic conversation device may add "XXX" to the survey "A", one of the morphemes, and XXY to the verb "A". But it is not limited thereto. Various methods can be used to distinguish phonemes, syllables, morphemes, or words that have the same sound but different meanings

The training data selector 230 included in the automatic conversation apparatus may select text to be used for model training. 2 illustrates the learning data selector 230 after the preprocessor 220, but is not limited thereto. After the operation of the training data selector 230 is performed, the operation of the preprocessor 220 may be performed.

The learning data selector 230 will be described in more detail with reference to FIG. 6.

6 illustrates a first text and a second text according to an embodiment of the present disclosure.

The automatic conversation device may receive free real-time consultation texts from professional counselors and users. The screen 600 is an example of free real-time consultation texts of a professional counselor and a user received by the automatic conversation device. The automatic conversation device may select text related to psychological counseling from real-time counseling text.

The automatic conversation apparatus may select texts related to psychological counseling using a previously trained learning model. For example, the previously trained learning model may be a model for learning conversation texts of a plurality of professional counselors. The automatic conversation apparatus may distinguish between the conversation text of the professional counselor and the conversation text of the professional counselor, based on the previously learned learning model. The automatic conversation apparatus may select a conversation text 620 of a professional counselor among real-time counseling texts based on a previously learned learning model.

In addition, the conversation texts 610 before the conversation text of the expert counselor may be selected together. The automatic conversation device may select a predetermined number of conversation texts 610. For example, the selected conversation texts may be three conversation texts 610. The conversation text of the selected professional counselor may be the second text 620. Also, the conversation texts before the conversation text of the expert counselor may be the first text 610.

The automatic dialog may automatically select text to be used for model training, but is not limited thereto. The automatic conversation device may receive text input from the user to use for model training. The automatic conversation apparatus may determine the text of the professional counselor described at the end of the input texts as the second text 620. Also, the text before the second text may be determined as the first text 610.

Hereinafter, the learning unit 110 of the automatic conversation apparatus will be described with reference to FIG. 7.

7 is a view illustrating a learning unit of an automatic conversation apparatus according to an embodiment of the present disclosure.

The model learner 710 included in the automatic conversation apparatus 700 uses a hierarchical encoder-decoder with attention mechanism to machine-learn a data set and machine-learning model 712. Operation 530 may be performed.

The hierarchical encoder-decoder model is a hierarchical learning model with a data set. The encoder represents a process of generating the machine learning model 712 based on the dialogue between the user and the professional counselor. The decoder represents the process of applying the user's dialogue to the machine learning model 712 to generate a natural response that will follow the user's dialogue.

The data included in the data set may be machine-learned and expressed as a vector or matrix such as {A1, A2, A3, A4, ... Ak}. A1, A2, A3, A4, ... Ak may each be a number. A1, A2, A3, A4, ... Ak may each have a real value. k means encoded in the k-dimensional. Representing data as vectors can be referred to as embedding. The expression of word units as vectors can be called word embedding. In addition, text embedding is a representation of text units as vectors.

The automatic conversation apparatus 700 may use an algorithm such as word2vec, fastText, GloVe, syllable unit word2vec, or Jamo unit word2vec for word embedding. In addition, the automatic conversation apparatus 700 may use the Word Mover's Distance (WMD) and doc2vec algorithms for text embedding.

The automatic conversation apparatus 700 may add a module called Long Short-Term Memory models (LSTM) or Gated Recurrent Unit (GRU) to Recurrent Neural Networks (RNN) for machine learning. LSTM and GRU are types of RNNs and algorithms for solving the long-term dependency problem of RNNs. In addition, one-way machine learning can be reversed to improve the basic model. In addition, one-way machine learning can be performed in both directions.

The automatic conversation apparatus 700 may perform machine learning in at least one layer of phonemes, syllables, morphemes, words, words, or sentences. Phonemes, syllables, morphemes, words, words, or sentences can each be hierarchical. In the phoneme, the syllable, the morpheme, the word, the word, and the sentence may be from the lower hierarchy unit to the higher hierarchy unit. A learning model trained using the upper layer can affect the learning model trained using the lower layer. On the contrary, a learning model trained using a lower layer may affect a learning model trained using a higher layer.

The automatic conversation apparatus 700 may perform a machine learning on the correlation between the upper hierarchical unit of the first text and the second text to obtain a weight for the upper hierarchical unit of the first text. The automatic conversation apparatus 700 may perform a machine learning on the correlation between the lower hierarchical unit of the first text and the second text to obtain a weight for the lower hierarchical unit of the first text. The automatic conversation apparatus 700 may perform a step of obtaining a weight for the modified lower layer unit based on the weight for the upper layer unit and the weight for the lower layer unit. The hierarchical unit may be a higher hierarchical unit from a lower hierarchical unit toward phonemes, syllables, morphemes, words, words, and sentences.

When acquiring the learning model only by the lower layer unit, the natural flow of the upper layer unit may not be reflected in the learning model. In addition, when acquiring the learning model only by the upper layer unit, the natural flow of the lower layer unit may not be reflected in the learning model. The automator can generate a natural response using a hierarchical encoder-decoder model.

Attention mechanism is a mechanism used for sequence to sequence based deep learning. Sequence-to-sequence deep learning is a technique for learning sentences using RNN (Recurrent Neural Network). Sequence-to-sequence deep learning has a disadvantage in that the longer the sequence does not remember the past state. Therefore, the attention mechanism aims to emphasize the part that the automatic communication device should pay attention to to obtain a result by using a separate network.

The learning unit 710 of the automatic conversation apparatus 700 may use the attention mechanism. The automatic conversation apparatus may learn the machine learning model 712 about the correlation between the first text and the second text. The automatic conversation apparatus 700 may pay attention to at least one of a phoneme, a syllable, a morpheme, a word, a word, or a sentence included in the first text in machine learning. This is because at least one unit of a phoneme, syllable, morpheme, word, word, or sentence included in the first text may be most relevant to the second text. For example, the first text may include the word "depressed", and the second text may be conversational text of a professional counselor who helps depression. The word "depressed" in the first text may have the highest impact on deriving the second text. Therefore, the learner 110 of the automatic conversation apparatus may generate the machine learning model 712 by focusing on the word “depressed” of the first text.

The recognition unit of the automatic conversation apparatus may use the attention mechanism. The automatic conversation device may store the machine learning model 712 received from the learner 710. The automatic conversation device may receive a conversation from the user. The conversation received from the user may correspond to the first text. The automatic conversation apparatus may generate the response text by applying the first text to the learning model. The response text may correspond to the second text. Since the learning model stored in the automatic conversation apparatus is a learning model generated by the attention mechanism, the second text may be generated by focusing on at least one of a phoneme, a syllable, a morpheme, a word, a word, or a sentence of the first text.

The automatic conversation apparatus 700 may perform a step of acquiring the machine learning model 712 about the correlation between the first text, the activity log of the user, and the psychological profile and the second text. The activity log of the user may include at least one of a post view, a comment, a sympathy, and a bookmark. The psychological profile may include at least one of a user's gender, age, disposition profile, and personality test results. Since the learning unit 710 of the automatic conversation apparatus 700 generates the machine learning model 712 by machine learning the activity log and the psychological profile of the user, the recognition unit of the automatic conversation apparatus 700 generates the machine learning model 712. Can be used to generate a response more appropriate to the user's situation.

8 is a flowchart illustrating a method of generating a response according to a dialogue of a user using a machine learning model according to an embodiment of the present disclosure.

The automatic conversation apparatus performs step 810 of receiving a first text related to psychoanalysis among conversation texts of the counselor and the user. The automatic conversation apparatus preprocesses the received first text to obtain a data set (820). The automatic conversation apparatus automatically obtains the second text from the received first text based on a previously learned machine learning model using a hierarchical encoder-decoder with attention mechanism. Step 830 is performed. The automatic dialogue device performs step 840 of outputting the second text.

9 is a block diagram of a recognition unit of an automatic conversation apparatus according to an embodiment of the present disclosure.

The automatic conversation apparatus 900 may include a recognition unit 910. The recognition unit 910 may be included in the same device as the learning unit 710. However, the present invention is not limited thereto, and the learner 710 may exist in a different device from the recognizer 910. The learner 710 may provide the machine learning model to the recognizer 910.

The recognizer 910 may include a data acquirer 310, a preprocessor 320, a recognition data selector 330, a recognition result provider 340, and a model updater 350 as described with reference to FIG. 3. Can be. The recognition unit 910 may receive the first text 911 and output the second text 912.

The data acquisition unit 310 included in the automatic conversation apparatus may perform step 810 of receiving a first text related to psychoanalysis among conversation texts of the counselor and the user. Additionally, the automated conversation device may receive a user's activity log and psychological profile. The activity log of the user may include at least one of a post view, a comment, a sympathy, and a bookmark. The psychological profile may include at least one of a user's gender, age, disposition profile, and personality test results.

The preprocessor 320 included in the automatic conversation apparatus may perform a step 820 of obtaining a data set by preprocessing the received first text. The automatic conversation apparatus may perform the step of separating the first text into at least one unit of phoneme, syllable, morpheme, word, word, or sentence.

The recognition data selector 330 included in the automatic conversation apparatus may select some text from the dialogue text from the user. The automator can select the last user's conversation text. The automatic conversation device may determine the conversation text of the user as the first text. In addition, the automatic conversation apparatus may set the dialogue text of the last user and the question text previously generated by the automatic dialogue apparatus as the first text.

The recognition result providing unit 340 included in the automatic conversation apparatus receives a first received based on a previously learned machine learning model using a hierarchical encoder-decoder with attention mechanism. Automatically acquiring the second text from the text may be performed 830.

10 is a view showing an operation screen of the automatic conversation apparatus according to one embodiment of the present disclosure.

The screen 1000 of the automatic communication device shows the first text. The first text may include conversation text 1011 of the automatic conversation device and conversation text 1012 of the user.

The automatic conversation device may receive a machine learning model learned in advance from the learning unit 110. The machine learning model may be a model generated based on a hierarchical encoder-decoder model with a mechanism of interest. The automatic conversation apparatus may generate the second text 1060 by applying the machine learning model to the first text 1011 and 1012. The second text 1060 may be conversation text that mimics the conversation of a professional counselor.

The automatic conversation apparatus may perform the step of automatically obtaining the second text 1060 from the received first text 1011, 1012, the activity log of the user, and the psychological profile based on the previously learned machine learning model. In addition to the first texts 1011 and 1012, the automator may generate a more natural second text 1060 using the user's activity log and psychological profile.

The machine learning model received by the automatic dialogue device may reflect a hierarchical encoder-decoder model with attention mechanism. The automatic conversation apparatus may perform the step of automatically obtaining the second text based on the weight of the upper layer unit, the weight of the lower layer unit, and the first text included in the previously learned machine learning model. The hierarchical unit may be a higher hierarchical unit from a lower hierarchical unit toward phonemes, syllables, morphemes, words, words, and sentences.

The automatic conversation apparatus may acquire the embedded first text by embedding the first text based on the previously learned machine learning model. As described above, the embedded first text may represent text in a vector or matrix format. The automatic conversation apparatus may perform the step of calculating the cosine similarity of the embedding of the plurality of candidates included in the embedded first text and the pre-learned machine learning model. The automatic conversation apparatus may include selecting a second text corresponding to the candidate most similar to the first text among the plurality of candidates based on the calculated similarity.

When the cosine similarity is smaller than the threshold, the automatic conversation device may perform the step of outputting a Polbeck response. The Polbeck response may be a response that prompts the user to enter conversation text again. Alternatively, the response may indicate that the user's conversation text cannot be processed. However, the present invention is not limited thereto, and when the dissimilarity level is higher than the threshold value, the automatic conversation apparatus may perform a step of outputting a Polbeck response.

So far I looked at the center of the various embodiments. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

Meanwhile, the above-described embodiments of the present invention can be written as a program that can be executed in a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. The computer-readable recording medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optical reading medium (eg, a CD-ROM, a DVD, etc.).

110: learning
120: recognition unit
210: data acquisition unit
220: preprocessing unit
230: training data selection unit
240: model learning unit
250: model evaluation unit

Claims (14)

In the automatic dialog for psychotherapy,
The apparatus includes a processor and a memory in which at least one program is stored,
The memory may be generated by the processor when the at least one program is executed in the processor.
Receiving texts related to psychoanalysis among conversation texts of a counselor and a user;
Preprocessing the received text for machine learning to obtain a data set; And
Storing instructions for executing the step of machine learning the data set to obtain a machine learning model using a hierarchical encoder-decoder with attention mechanism;
The received text includes a first text including a conversation text of the counselor and the user and a second text that is a response text of the counselor immediately after the first text,
Acquiring the machine learning model,
Obtaining a machine learning model for the correlation between the first text and the second text.
The method of claim 1,
Instructions for causing the processor to execute the step of obtaining the machine learning model,
Obtaining a machine learning model for the correlation of the first text, the activity log of the user, and the psychological profile and the second text,
The activity log of the user includes at least one of viewing posts, comments, empathy and bookmarks,
And the psychological profile comprises at least one of a gender, age, disposition profile, and personality test result of the user.
The method of claim 1,
Instructions for causing the processor to execute the step of acquiring the data set include:
Dividing the received text into a first text and a second text; And
And dividing the first text and the second text into at least one unit of phoneme, syllable, morpheme, word, word or sentence.
The method of claim 1,
Instructions for causing the processor to execute the step of obtaining the machine learning model,
Performing machine learning on the correlation between the higher layer unit of the first text and the second text to obtain a weight for the higher layer unit of the first text;
Performing machine learning the correlation between the lower hierarchical units of the first text and the second text to obtain weights for the lower hierarchical units of the first text; And
Obtaining a weight for the modified lower layer unit based on the weight for the upper layer unit and the weight for the lower layer unit,
The hierarchical unit is a phone, syllables, morphemes, words, words and sentences, the automatic dialogue device characterized in that the lower hierarchical unit from the upper hierarchical unit.
In the automatic dialog for psychotherapy,
The apparatus includes a processor and a memory in which at least one program is stored,
The memory may be generated by the processor when the at least one program is executed in the processor.
Receiving a first text related to psychoanalysis among conversation texts of a counselor and a user;
Preprocessing the received first text to obtain a data set;
Automatically obtaining a second text from the received first text based on a previously learned machine learning model using a hierarchical encoder-decoder with attention mechanism; And
And save instructions to execute the step of outputting the second text.
The method of claim 5,
Instructions for causing the processor to execute the step of automatically obtaining the second text include:
Automatically obtaining a second text from the received first text, an activity log of the user, and a psychological profile based on the pre-learned machine learning model,
The activity log of the user includes at least one of viewing posts, comments, empathy and bookmarks,
And the psychological profile comprises at least one of a gender, age, disposition profile, and personality test result of the user.
The method of claim 5,
Instructions for causing the processor to execute the step of acquiring the data set include:
And dividing the first text into at least one unit of phoneme, syllable, morpheme, word, word, or sentence.
The method of claim 5,
Instructions for causing the processor to execute the step of automatically obtaining the second text include:
Automatically obtaining a second text based on a weight for a higher layer unit, a weight for a lower layer unit, and the first text included in the pre-learned machine learning model,
The hierarchical unit is a phone, syllables, morphemes, words, words and sentences, the automatic dialogue device characterized in that the lower hierarchical unit from the upper hierarchical unit.
The method of claim 5,
Instructions for causing the processor to execute the step of automatically obtaining the second text include:
Embedding the first text based on the pre-learned machine learning model to obtain an embedded first text;
Calculating a cosine similarity between embedding of the embedded first text and a plurality of candidates included in the pre-learned machine learning model; And
And selecting a second text corresponding to a candidate most similar to a first text among the plurality of candidates based on the calculated similarity.
The method of claim 8,
The memory may be generated by the processor when the at least one program is executed in the processor.
An automatic conversation device storing instructions for executing a step of outputting a polbeck response when the cosine similarity is smaller than a threshold value
In the automatic dialogue method for psychotherapy,
Receiving texts related to psychoanalysis among conversation texts of a counselor and a user;
Preprocessing the received text for machine learning to obtain a data set; And
Machine learning the data set using a hierarchical encoder-decoder with attention mechanism to obtain a machine learning model,
The received text includes a first text including a conversation text of the counselor and the user and a second text that is a response text of the counselor immediately after the first text,
Acquiring the machine learning model,
Obtaining a machine learning model for the correlation between the first text and the second text.
In the automatic dialogue method for psychotherapy,
Receiving a first text related to psychoanalysis among conversation texts of a counselor and a user;
Preprocessing the received first text to obtain a data set;
Automatically obtaining a second text from the received first text based on a previously learned machine learning model using a hierarchical encoder-decoder with attention mechanism; And
And outputting the second text.
A computer program product comprising a computer readable recording medium having recorded thereon a program for executing the method of claim 11 on a computer.
A computer program product comprising a computer readable recording medium having recorded thereon a program for executing the method of claim 12 on a computer.

Images