KR102263656B1 - Communication system for inducing response and communication method for inducing response - Google Patents

Abstract

A communication system comprises: a user interface configured to receive an utterance from a user of a communication system and provide the utterance of the communication system to the user; a natural language processor configured to process the user's utterance and convert the utterance into user utterance data for which the communication system can interpret; a database configured to store at least one communication domain, wherein the communication domain comprises a system utterance data; and an artificial intelligence module configured to generate the utterance of the communication system by responding to receiving the utterance from the user and referencing to the natural language processor and the database. The communication system is configured to draw the user to utter a particular target. By drawing the user to utter a plurality of sub-targets prior to drawing the user to utter speak the specific target, the artificial intelligence module is configured to set a degree of similarity between a sub-target relatively close to the specific target and the specific target to be greater than or equal to a similarity between a sub-target relatively distant to the specific target and the specific target. Therefore, the present invention is capable of allowing the user to feel as if he/she is communicating with a real person.

Description

Conversational systems and methods of inducing answers {COMMUNICATION SYSTEM FOR INDUCING RESPONSE AND COMMUNICATION METHOD FOR INDUCING RESPONSE}

The present disclosure relates to a dialog system, and more particularly, to a dialog system that induces a desired answer.

The dialogue system is a system that sends a response to the user's remarks. The chat system of a specific domain provides a chat environment to users who visit the chat system with a specific goal. For example, a user who wants to consult for insurance subscription may receive a response to an inquiry using the insurance domain dialogue system. A dialog system of a specific domain is also called a close domain because a user inputs a specific keyword with a specific goal from the beginning and the dialog system responds to it. In the case of the closed domain, since it is learned using specific keywords, only general and monotonous conversations are possible, and it is difficult to conduct conversations like a real person. The following prior art documents are incorporated herein by reference in their entirety and may be used to describe the disclosure of the present specification.

Korean Patent Registration No. 10-2008169 Korean Patent Publication No. 10-2198295

Representing long texts for NLU, CS224U Stanford University

The present disclosure provides a dialog system for performing a conversation by understanding the context of the conversation through questions and answers. Accordingly, the conversation system may provide a conversation in which the user feels as if he or she is having a conversation with a real person and may also elicit a desired answer.

According to one aspect of the present disclosure, an interactive system is provided. The conversation system includes: a user interface configured to receive a utterance from a user of the conversation system and provide the utterance of the conversation system to the user; a natural language processor configured to process the user's speech and convert it into user speech data that the dialog system can interpret; a database configured to store at least one conversation domain, the conversation domain including system speech data; and an artificial intelligence module configured to generate a utterance of the dialog system with reference to the natural language processor and the database in response to receiving a utterance from the user. The conversation system is configured to induce the user to utter a specific goal. The artificial intelligence module induces the user to speak a plurality of sub-goals before inducing the user to speak the specific target, and the artificial intelligence module is configured to include sub-goals relatively close to the specific target and the and set such that the similarity of the specific target is greater than or equal to the similarity between the sub-goal relatively distant from the specific target and the specific target.

According to an embodiment, the artificial intelligence module generates and provides an i-th utterance of the dialog system in response to the i-1 th utterance of the user, wherein each of one or more utterances among the utterances before the ith utterance of the user and the i-th utterance of the conversation system is generated based on the degree of similarity to the specific target and the similarity between the i-th utterance that the user is expected to make in response to the i-th utterance of the dialogue system and the specific target. can do.

According to an embodiment, the artificial intelligence module is configured to correspond to the i-th remark of the dialog system, and the degree of similarity between the i-th remark of the user expected to speak by the user and the specific target, and the k-th remark of the user from i - to generate the i-th utterance such that the average value of the sum of the similarities between each of the -1 th utterances and the specific target is equal to or greater than the average value of the similarity between each of the i-1 th utterances of the user and the specific target can Wherein k and l are natural numbers less than i-1 and greater than 1.

According to an embodiment, the conversation system is configured to count the conversation of the i-1 time, and after counting the i-1 time, generate and provide the i-th utterance in response to the user's i-1 utterance can be

According to an embodiment, the database is configured to store a plurality of dialogue domains, each of the plurality of dialogue domains is a dialogue domain configured to derive a different specific goal, and the dialogue system is any one of the plurality of dialogue domains. can be configured to select

According to one aspect of the present disclosure, there is provided a conversation method of a conversation system that induces a user to speak a specific target. The conversation system includes: a user interface configured to receive a utterance from a user of the conversation system and provide the utterance of the conversation system to the user; a natural language processor configured to process the user's speech and convert it into user speech data that the dialog system can interpret; a database configured to store at least one conversation domain, the conversation domain including system speech data; and an artificial intelligence module configured to generate a utterance of the dialog system with reference to the natural language processor and the database in response to receiving a utterance from the user. The conversation method may include, before inducing the user to speak the specific target, inducing the user to speak a first sub-goal; inducing the user to speak a second sub-goal after inducing the user to speak the first sub-goal and before inducing the user to speak the specific target; and inducing the specific target to be spoken. The similarity between the second sub-goal and the specific target is equal to or greater than the similarity between the first sub-goal and the specific target.

In one embodiment, the artificial intelligence module further comprises the step of generating and providing the i-th utterance of the dialog system in response to the i-1 th utterance of the user. The step of generating and providing the ith utterance of the dialog system may include: a degree of similarity between one or more utterances of the user's utterances before the ith utterance to the specific target, and the user in response to the ith utterance of the dialog system. and generating the i-th utterance of the dialogue system based on the similarity between the ith utterance expected to be uttered and the specific target.

In one embodiment, the generating and providing the i-th utterance of the conversation system includes the ith utterance of the user expected to speak in response to the i-th utterance of the conversation system and the specific target. calculating a first average value that is an average value of the similarity and the sum of the user's k-th utterances to the i-1 th utterances and the similarity of the specific target; calculating a second average value that is an average value of the similarity between each of the i-1 th utterances from the 1 th utterance of the user and the specific target; and generating the i-th utterance such that the first average value is equal to or greater than the second average value. Wherein k and l are natural numbers less than i-1 and greater than 1.

In one embodiment, the dialog system counts the conversation i-1, and after counting the i-1, generating the utterance i in response to the utterance i-1 of the user. may further include.

In an embodiment, the database is configured to store a plurality of conversation domains, each of the plurality of conversation domains may be a conversation domain configured to derive a different specific goal. The conversation method may further include selecting any one of the plurality of conversation domains before inducing the first bus target to be spoken.

The conversation system can provide a conversation that makes the user feel as if he or she is talking to a real person and can also elicit the desired answer.

1 is a block diagram of a conversation system according to an embodiment;
2 is an embodiment of a dialogue provided by a dialogue system according to the present disclosure to induce the word “I want to invest”.
3 is a conceptual diagram for explaining a topic switching method according to the present disclosure.
4 and 5 are an example of a conversation between a conversation system and a user according to an embodiment of the present disclosure.
6 is a flowchart illustrating a method of deriving a final goal according to an embodiment of the present disclosure.

Hereinafter, the present disclosure will be described in detail with reference to the drawings. In describing the present disclosure, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present disclosure, a detailed description thereof will be omitted. In addition, the following examples may be modified in various other forms, and the scope of the technical spirit of the present disclosure is not limited to the following examples. Rather, these embodiments are provided to more fully and complete the present disclosure, and to fully convey the technical spirit of the present disclosure to those skilled in the art.

It is to be understood that the techniques described in the present disclosure are not intended to be limited to specific embodiments, and include various modifications, equivalents, and/or alternatives of the embodiments of the present disclosure.

In connection with the description of the drawings, like reference numerals may be used for like components.

In the present disclosure, expressions such as “have,” “may have,” “include,” or “may include” indicate the presence of a corresponding characteristic (eg, a numerical value, function, operation, or component such as a part). and does not exclude the presence of additional features. In this disclosure, expressions such as “A or B,” “at least one of A and/and B,” or “one or more of A or/and B” may include all possible combinations of the items listed together. . For example, "A or B," "at least one of A and B," or "at least one of A or B" means (1) includes at least one A, (2) includes at least one B; Or (3) it may refer to all cases including both at least one A and at least one B.

As used in the present disclosure, expressions such as "first," "second," "first," or "second," may modify various elements, regardless of order and/or importance, and refer to one element. It is used only to distinguish it from other components, and does not limit the components.

A component (eg, a first component) is "coupled with/to (operatively or communicatively)" to another component (eg, a second component) When referring to "connected to", it will be understood that the certain element may be directly connected to the other element or may be connected through another element (eg, a third element). On the other hand, when it is said that a component (eg, a first component) is "directly connected" or "directly connected" to another component (eg, a second component), the component and the It may be understood that other components (eg, a third component) do not exist between other components.

The expression "configured to (or configured to)" as used in this disclosure depends on the context, for example, "suitable for," "having the capacity to" ," "designed to," "adapted to," "made to," or "capable of." The term “configured (or configured to)” may not necessarily mean only “specifically designed to” hardware. Instead, in some circumstances, the expression “a device configured to” may mean that the device is “capable of” with other devices or parts. For example, the phrase "a processor configured (or configured to perform) A, B, and C," a module configured (or configured to perform) A, B, and C," refers to a dedicated processor (e.g., : embedded processor) or a generic-purpose processor (eg, CPU or application processor) capable of performing corresponding operations by executing one or more software programs stored in a memory device.

dialogue system

1 is a block diagram of a conversation system according to an embodiment;

Referring to FIG. 1 , the dialog system 10 includes a user interface (UI, 100), a natural language processor (NLP, 200), a processor 300, a database (DB, 400), and an artificial intelligence module (AI Module, 500). include

The user interface 100 receives a utterance from the user 1 through a network (not shown) and provides the utterance to the user 1 . The user interface 100 may provide an interface for accessing the chat system 10 through a terminal used by the user 1 . The user 1 (10) may transmit the user 1 (10) utterance to the conversation system 10 through the user interface 100, and the user 1 (10) can transmit the utterance from the conversation system 10 through the user interface 100. can be delivered An utterance is a sentence that the user 1 or the conversation system 10 intends to convey to the other party in the course of a conversation.

The communication method of the network is GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), HSDPA (High Speed Downlink Packet Access), HSUPA (High Speed Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced, etc.), WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX ( A network constructed according to World Interoperability for Microwave Access) may be used, but the present invention is not limited thereto and may include all transmission method standards to be developed in the future. It may include anything that can send and receive data through wired/wireless.

The natural language processor 200 is configured to perform natural language processing on the user's 1 utterance. As will be described later, the natural language processor 200 may be a language model obtained by learning a learning model including an artificial neural network (ANN). For example, the natural language processor 200 is Google's BERT ((Bidirectional Encoder Representation from Transformers), and a model applying it), GPT ((Generative Pre-Training), and a model applying it), XLNET, RoBERTa, ALBERT and the like.

The natural language processor 200 may perform morpheme analysis, entity name recognition, syntax analysis, sentiment analysis, semantic analysis, intention analysis, lexical semantic analysis, semantic translation analysis, pragmatic analysis, information search, sentence semantic analysis, and the like.

For example, natural language processing is a morpheme analysis process that analyzes a target word into a morpheme that is the smallest unit of meaning, and a process of selecting the most suitable form of speech among the results of the morpheme analyzed in this way. Information serving as a hint for disambiguation located on the left and right of the context A part-of-speech attachment process that selects an appropriate analysis result using , a phrase-based analysis process that analyzes lumps such as noun phrases, verb phrases, and adverb phrases, a clause-unit analysis process that decomposes sentences such as Chinese and complex sentences into short sentences, and It may include a syntax analysis process of determining the structure of the sentence by analyzing the hierarchical relationship between the constituent elements.

Using the above-described language model, it is possible to classify natural languages, extract features, and vectorize them. Simply put, natural language is a language that people use in their daily life. Since the dialog system 10 is a machine, the natural language processor 200 is configured to transform the dialog system 10 so that it can decode the natural language and interpret its meaning. In an embodiment, the natural language processor 200 categorizes human speech into data. That is, the natural language processor 200 may embed the sentence or word input by the user 1 and convert it into data (user speech data) expressed in an n-dimensional vector format. Hereinafter, the embedded sentence or word input by the user 1 is referred to as a user speech vector or user speech data. Natural language processing can be easily performed by a person of ordinary skill in the art with reference to a document described as a prior document in the present disclosure, and thus a detailed description thereof will be omitted. However, in the present disclosure, natural language processing is not limited to the contents of prior documents, and a range that can be implemented by a person of ordinary skill in the art is included in the present disclosure.

The processor 300 is configured to manage and control components of the conversation system 10 to conduct a conversation between the user 1 and the conversation system 10 . For example, the processor 300 is ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), controllers (controllers) ), micro-controllers, microprocessors, or any other type of processor or controller for performing other functions.

Database 400 is configured to store information for operation of dialog system 10 . The database 400 may store a plurality of application programs driven in the dialog system 10 or data and instructions that can be read by the processor 140 of the application dialog system 10 . For example, the dialogue system 10 may be configured in various ways such as a hard disk drive (HDD), a solid state disk (SSD), a silicon disk drive (SDD), ROM, RAM, EPROM, flash drive, hard drive, cloud using a network, and the like. It may include storage space.

In one embodiment, the database 400 may include a plurality of conversation domains, data for creating conversations, and the like. The plurality of conversation domains may be a conversation model configured to fit a conversation situation. Each of the plurality of conversation domains may be a conversation model corresponding to a final goal. The dialogue domain of the database 400 may include a plurality of sentences or data in an n-dimensional vector format (hereinafter, system speech vector or system speech data) corresponding to each of the plurality of sentences. The system speech vector has been described as n-dimensional vector format data corresponding to a plurality of sentences or each of the plurality of sentences included in the dialogue domain of the database 400 , but the artificial intelligence module 500 works with the natural language processor 200 . It may include data obtained by converting a natural language generated by operation or predicted to be generated into data in an n-dimensional vector format.

The artificial intelligence module 500 includes an artificial intelligence model trained and configured to perform a conversation in response to the user 1's utterance. An artificial intelligence model is, in the present disclosure, by learning a learning model including an artificial neural network (ANN) through a large amount of learning data to optimize parameters inside the artificial neural network, and a conversation system using the learned learning model It may mean a model involved in the operation of (10). In an embodiment, the artificial intelligence module 500 may be learned through Machine Reading Comprehension (MRC). In one embodiment, the artificial neural network model used in the artificial intelligence module 500 is a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), limited At least one of a Restricted Boltzmann Machine (RBM), a Deep Belief Network (DBN), a Bidirectional Recurrent Deep Neural Network (BRDNN), or Deep Q-Networks, or There are combinations of these, but it is not limited to the above-described examples.

The artificial intelligence module 500 may perform context maintenance, query determination, optimal correct answer generation, correct answer candidate inference, correct correct candidate based inference, correct correct candidate generation, apex reliability inference, correct answer type analysis, correct answer constraint recognition, and question answer knowledge learning. have. The artificial intelligence module 500 may manage the dialogue of the dialogue system 10 , analyze a sentence relationship, and perform learning through feedback.

The artificial intelligence module 500 may generate the utterance of the dialog system 10 in natural language with reference to the natural language processor 200 and/or the database 400, and the utterance of the dialog system 10 generated in the natural language is processed by the processor. It may be transmitted to the user 1 through the user interface 100 by means of 300 . Since the dialogue system 10 generates a utterance can be easily performed by a person skilled in the art with reference to a document described as a prior document in the present disclosure, a detailed description thereof will be omitted. However, the generation of utterances in the present disclosure is not limited to the contents of prior documents, and a range that can be implemented by a person of ordinary skill in the art is included in the present disclosure.

In an embodiment, the artificial intelligence module 500 may compare the similarity between the user speech vector and the system speech vector, and use the result to generate or select the speech of the dialog system 10 . The artificial intelligence module 500 may select at least one conversation domain stored in the database 400 and generate a utterance of the conversation system 10 using the selection. The artificial intelligence module 500 may select at least one conversation domain based on the final goal.

1 is an example in which the configuration of the dialog system 10 is separated for explanation, each component can perform the functions described above using a combination/integration or other components, and a description of each component is provided. It can be applied to other configurations. For example, the natural language processor 200 may be combined with or integrated with the artificial intelligence module 500 , or may use the artificial intelligence module 500 . The artificial intelligence module 500 may use the database 400 and/or the natural language processor 200 . The processor 300 may use the natural language processor 200 in combination with the artificial intelligence module 500 , or may be integrated with the artificial intelligence module 500 .

Learning the Conversational System

The learning data collection for learning of the conversation system 10 is 1) collecting existing chatbot or agent chat conversations of financial companies, 2) collecting phone consultation history (including voice recording), 3) collecting comments for conversations, 4) salespersons (agents) ) and the interviewee (user (1)) collection of consultation contents, 5) virtual consultation contents collection, and the like.

The collected data is refined/processed to make learning data. In the case of voice data, the data is converted through voice recognition. Synonyms, neologisms, and substitutes are organized to classify and process data according to the context of the conversation.

The natural language processor 100 and/or the artificial intelligence module 500 is trained using the learning data. Accordingly, a database 400 related to conversation cases is built. The learned natural language processor 100 and/or artificial intelligence module 500 is re-learned based on conversational feedback.

By learning, the natural language processor 100 and/or the artificial intelligence module 500 of the conversation system 10 may generate a utterance in response to the utterance of the user 1 and provide it to the user 1 .

Behavior of the dialogue system

Basically, the dialog system 10 is configured to provide a response in response to a utterance of the user 1 . The dialog system 10 is configured, in more detail, such that the user 1 utters the dialog system 10 or a goal intended by the user 1 of the dialog system 10 .

For example, the dialog system 10 may aim to induce the user 1 to say “I want to invest”. To this end, the conversation system 10 may induce the user 1 to naturally say “I want to invest” as if in a conversation with a person.

In the case of a conversation between the interviewee (user) and the agent (person), at the beginning of the conversation, or immediately after starting the conversation, goal-directing remarks (for example, a question like Do you want to invest?) are sent by the agent to the user (1 ) is rarely spoken. This is because the user 1 may feel unpleasant when a target-inducing remark is made at the same time as starting a conversation or immediately after starting a conversation.

Therefore, in general, the counselor has a daily conversation with the user 1 before inducing the final target statement, or forms intimacy through surrounding stories, then changes the direction of the conversation to the next topic, and changes the direction of the conversation. You can take at least one step and then lead to the final target statement. For example, if the counselor and the user 1 do not know anything at all, after greetings, they can start a conversation with topics unrelated to their goals, such as general society, weather, entertainment, sports, the world, and science. Then, the counselor changes the topic to an issue related to investment or economy, for example, a socially known economic issue (for example, the salary story of a celebrity or sports player), and from this, the user's (1) interest can be turned into an economic issue. Then, the counselor starts a conversation with an economic issue, tries to have a conversation to extract information related to the user (1) (investment propensity, asset size, economic knowledge, needs of the user (1), etc.), and then switches to the next topic. can be prepared For example, the next topic may include examples of various investments, information on currently popular investments, and the like.

2 is an embodiment of a dialogue provided by the dialogue system 10 according to the present disclosure to induce the word “I want to invest”.

Referring to FIG. 2 , as a human counselor induces a statement that he wants to invest in the user 1 , the conversation system 10 may provide at least one topic switching step for changing the direction of the conversation. The topic shift phase is the phase provided between the beginning of a conversation and the final goal (ie, an indication of a desire to make an investment). The topic switching phase includes each sub-goal G1, G2, G3. The sub-goals G1, G2, and G3 are the goals of the topic change phase. That is, the conversation system 10 may induce the user 1 to speak each sub-goal G1 , G2 , and G3 , and finally to speak the final target.

In the daily conversation phase, the dialogue system 10 may include, for example, a dialogue generation model implemented based on GPT-3. The daily conversation step may include a step of providing the user 1 with topics such as greetings, recent society, life, culture, world, politics, science, and economy in response to the user 1 's remarks.

After exchanging a plurality of conversations in the daily conversation phase, the conversation system 10 may start a conversation to change the topic.

In one embodiment, the degree of similarity between the user 1's remarks or the dialogue system 10's remarks and the final goal in the topic switching phase is the same as the user 1's remarks or the dialogue system 10's remarks in the daily conversation phase. greater than the similarity to the final goal. In addition, the degree of similarity between the user 1's remarks or the remarks of the conversation system 10 and the final goal in the topic switching step 2 is the user 1's remarks or the remarks of the dialogue system 10 and the final goal in the topic switching step 1 greater than the similarity to Accordingly, in the daily conversation phase, as the final goal such as the topic switching steps 1, 2, and 3 is approached, the similarity between the user 1's statement or the conversation system 10's statement and the final goal may increase.

In one embodiment, the degree of similarity between the sub-goal and the final goal in the step of switching the topic is greater than the degree of similarity between the speech of the user 1 or the speech of the conversation system 10 and the final goal in the step of switching the topic in the daily conversation step. The similarity between the sub-goal G2 and the final goal in the topic change step 2 is greater than the similarity between the sub-goal G1 and the final goal in the topic change step 1 . The similarity between the sub-goal G3 and the final goal in the topic change step 3 is greater than the similarity between the sub-goal G2 and the final goal in the topic change step 2 . As described above, in the daily conversation stage, the similarity between the sub-goal and the final goal may increase as the final goal, such as the topic switching steps 1, 2, and 3, approaches.

3 is a conceptual diagram for explaining a topic switching method according to the present disclosure.

Referring to FIG. 3 , “G” is the goal of the topic conversion stage (the final goal, for example, the sub-goal of the intermediate stage for inducing ‘I want to invest’, hereinafter the sub-goal), and “H” is the person The utterance of “C” is the utterance of the dialogue system 10, and the numbers next to “H” and “C” indicate the order of the utterances. For example, H ₀ is the utterance provided to the conversation system 10 when the user 1 uses the conversation system 10 for the first time, C ₁ is the first utterance of the conversation system 10, and H ₁ is C _It is the remark of the user 1 corresponding to 1. That is, H _n may be a statement corresponding to C _n (n is a positive integer).

The topic switching step is a step for inducing a sub-goal from the user 1 . Accordingly, in the topic switching phase, the dialogue system 10 is configured to determine which utterance to provide in consideration of the sub-goal or based on the sub-goal.

In one embodiment, the conversation system 10 generates a utterance in consideration of a similarity between the sub-goal G and the utterance provided to the user 1 and/or the similarity between the sub-goal G and the user 1 utterance. can be provided. For example, when the dialog system 10 provides the i-th utterance, the dialog system 10 provides at least one of the sub-target G and the utterance provided to the user 1 or the sub-goal G and the user ( A statement may be generated and provided in consideration of the degree of similarity of at least one of the statements of 1).

In one embodiment, when the dialog system 10 provides the i-th utterance, the dialog system 10 considers the similarity of the utterance of the dialog system 10 to the utterance that the user 1 is expected to utter in response thereto. You can create a statement and provide it.

In one embodiment, when the dialog system 10 provides the ith utterance, the dialog system 10 determines the similarity between the utterance of the dialog system 10 before the ith utterance and the sub-goal G, the sub-goal G It is possible to generate and provide a remark in consideration of the similarity of remarks expected to be remarked by the user 1 in response to and i-th remarks.

In an embodiment, the similarity may include a cosine similarity.

For example, the conversation system 10 may determine the i-th remark after exchanging a plurality of conversations based on the similarity between the plurality of conversations and the sub-goal G. For example, the dialogue system 10 may determine the ith utterance using Equation 1 below.

Here, k and l are the number of utterances (an integer greater than or equal to 1) considered by the dialogue system 10, and i is the number of utterances. Accordingly, the conversation system 10 is configured to count whether the conversation system 10 and the user 1 have performed the conversation i-1 or more times. After the user 1 and the dialog system 10 have a conversation multiple times or more, the conversation system 10 may apply Formula 1 to perform the conversation. According to Equation 1, the dialogue system (10) i-th speech (C _i) on the expected user speech (H _i) and the similarity of the sub-target and the i-th speech plurality before (C _i) one (for example, above about In the example k), the average value of the similarity between the user's statement and the sub-goal is greater than or equal to the average value of the similarity between the user's statements and the sub-goal of a plurality of users before the i-th statement (for example, l in the example above). The second utterance is selected. k and l are configurable. For example, if we consider 5 utterances (k=l=5), Equation 1 becomes the following Equation 2.

Assuming that the 10th utterance (i=10) of the dialogue system is obtained by considering 5 user utterances, the above formula becomes the following formula 3.

That is, the similarity between each of the 5th to 9th remarks of the user 1 and the sub-goal and the 10th of the dialogue system 10 is higher than the average value of the similarity between each of the user 1's 4th to 9th remarks and the sub-goal. The tenth utterance of the dialog system 10 is selected such that the average of the sum of the degree of similarity between the user's tenth utterance and the sub-goal expected for the utterance is equal to or greater than the average.

recommended Cn)은, 챗봇의 발언에 대해 예상되는 사람의 발언의 임베딩 값(벡터값)이다. 상술한 것과 같이 학습된 대화 시스템(10)은 사용자(1)에 대해 발언을 생성할 수 있다. 일 실시예에서, 사용자(10)의 발언에 대응한 대화 시스템(10)의 발언의 후보는 복수 개가 있을 수 있다. 또한, 대화 시스템(10)의 발언의 후보에 대해 예상되는 사용자(10)의 발언의 후보는 복수 개가 있을 수 있다. 이러한 다양한 가능성을 서브 목표와의 유사도를 계산하여 상기 공식에 맞추어 대화 시스템(10)의 i번째 발언이 선택될 수 있다. (predicted Hn

recommended Cn) is an embedding value (vector value) of a person's speech expected for the chatbot's speech. As described above, the learned conversation system 10 may generate a utterance for the user 1 . In an embodiment, there may be a plurality of candidates for the speech of the conversation system 10 corresponding to the speech of the user 10 . In addition, there may be a plurality of candidates for the speech of the user 10 expected for the candidates for the speech of the conversation system 10 . The i-th utterance of the dialogue system 10 may be selected according to the above formula by calculating the degree of similarity to the sub-goal for these various possibilities.

In one embodiment, the dialog system 10 predicts the user's i-th utterance and selects the utterance in order to induce the sub-goal G in the topic switching step. In more detail, the dialog system 10 is configured to induce the user's i-th utterance, and the similarity between at least one of the user's utterances before the i-th utterance and the sub-goal G and the user corresponding to the i-th utterance of the dialog system 10 . The i-th remark of the dialogue system 10 may be selected in consideration of the similarity between the i-th remark of and the sub-target G.

The method of the dialogue system 10 may be variously modified. (The following is a modified example of the left side of the equation.) In an embodiment, the method of the dialogue system 10 may select the i-th dialogue using at least one of the following formulas.

Here, j, i, k, and l are an integer greater than or equal to 0 or an integer greater than or equal to 1, and i is the number of utterances.

On the right side of the formula, _{the similarity between the user 1's utterance before the i-th utterance (H i} ) and the sub-goal (G) or at least a portion of the utterances (H ₁ ... H _{1-i ) before the ith utterance (H i )} represents the average of the degree of similarity between each of the sub-objectives (G).

The left side of the formula contains the similarity between the i-th utterance (H _i ) of the user 1 and the sub-goal (G), and the utterances before the _{i-th utterance (H i ) (H 1} ...H _1-i ) or This is a value calculated in consideration of the similarity between the predicted utterance after the i-th utterance (H _{i ) and the sub-goal (G).} Alternatively, the left side of the formula may be a value obtained by estimating the user 1's speech as a whole, obtaining a similarity to each sub-target, and obtaining an average value thereof.

In one embodiment j, i, k, l may be counted by the dialog system 10 . Since the above formulas can be applied after more than j, i, k, l or more conversations have been performed, the conversation system 10 performs a plurality of conversations between the user 1 and the conversation system 10, and then the j, i After determining whether the number of times , k, and l is satisfied, the above formula may be applied to induce the user 1 to reach the sub-goal.

The above formula is one of various modifications, and in order to induce the sub-goal by inducing the i-th remark of the user 1 as in the present disclosure, the sub-goal and the i-th of the user 1 and/or the dialog system 10 Using the similarity between the previous statement and the sub-goal, the above formulas can be variously modified. For example, the dialog system 10 may use i+1, i+2, ..., i+p (p is 1) of the dialog system 10 and/or user 1 to generate an i-th utterance. more than a natural number), a utterance may be predicted, a degree of similarity between the predicted utterance and the sub-goal may be calculated, and an i-th utterance may be generated based thereon.

In one embodiment, since the dialog system 10 checks the similarity between the user's utterance and the sub-goal G, the dialog system 10 can determine whether the user's utterance is approaching the sub-goal G, and can be used to select utterances of the dialogue system 10 . Dialog system 10 is to induce a user in the i-th speech (H _i), the user of the i-th speech (H _i) and the i-th speech of the degree of similarity is the user of the sub-goals (G) speech before (H _i) ( H ₁ ,...H _1-i ) and the sub-goal (G) or the user's i-th utterance (H _i ) before the utterance (H _k ,...H _1-i ) and the sub-goal (G) _{The i-th utterance (C i} ) may be selected to be equal to or greater than the average of the similarity.

As described with reference to FIGS. 2 and 3 , the conversation system 10 may change the direction of the conversation by inducing the user to make a desired remark by the conversation system 10 in the step of switching the topic. Accordingly, the conversation system 10 may induce the user to speak a sub-goal of the topic change step.

4 and 5 are an example of a conversation between the conversation system 10 and the user 1 according to an embodiment of the present disclosure. 4 and 5 are only for showing an embodiment of the dialog system 10 of the present disclosure in a virtual embodiment, and the conversation system 10 of the present disclosure is not limited thereto.

The method of selecting the i-th utterance of the dialogue system to induce a sub-goal in the topic switching step of the dialogue system 10 described with reference to FIG. 3 may be applied to the method of selecting the final target of FIG. 2 . For example, the sub-goal of the topic change step may be the same as the final goal. In this case, the conversation system 10 may induce the user 1 to speak the final goal by conducting a plurality of conversations and inducing i-th, i+1-th, and i+2-th utterances.

A method of selecting a speech by the dialogue system 10 in consideration of a similarity between the target and the remarks of the user 1 or the remarks of the dialogue system 10 can be variously applied.

6 is a flowchart illustrating a method of deriving a final goal according to an embodiment of the present disclosure.

In one embodiment, the method described with reference to FIG. 3 and the above-described formulas may be applied to determine the topic change step or the sub-goal of the topic change step. That is, it is possible to create a scenario in which the final goal is derived by setting the topic switching step based on the degree of similarity between the final goal and the sub-goal. Then, a utterance that can induce a utterance of the user 1 is selected using the description and formulas of the present disclosure to derive a sub-goal of each topic change step.

Referring to FIG. 6 , the dialogue system 10 obtains a final goal ( S605 ). Achieving the final goal means that the user of the conversation system 10 inputs the final goal into the conversation system 10 so that the conversation system 10 receives the final goal, any one of the final goal stored in the conversation system 10 the user 1 accesses the dialogue system 10 and inputs a sentence or word related to the end goal so that the dialogue system 10 infers the end goal therefrom; ) and entering the final target related sentences or words, it may include things that are set in advance.

The dialogue system 10 sets a dialogue scenario based on the final goal (S610). The dialogue system 10 may set up a dialogue scenario including at least one topic switching step. The topic shift phase includes sub-goals. The sub-goal is a speech that the dialog system 10 intends to induce the user 1 to speak. In one embodiment, when there are a plurality of topic conversion steps, the degree of similarity between the final target and the sub-goal of the topic switching step relatively close to the final goal is similar to the similarity between the sub-goal and the final goal of the topic conversion step relatively far from the final goal. Conversation scenarios may be set to be equal or greater.

The dialogue system 10 performs dialogue according to the set scenario (S615). The model for inducing the user 1's speech described with reference to FIGS. 2 and 3 may be applied to the conversation scenario.

In one embodiment, the above-described final goal “I want to invest” is only an example, and the final goal may include other topics or be modified. For example, the end goal is set by the user 1 using the conversation system 10, such as “I want to buy insurance,” “I want to buy a product,” “I want to subscribe to a service,” or the conversation system 10 ) may be at least one of a number of pre-stored goals.

In an embodiment, the method of predicting and inducing a speech using a similarity with the sub-goal to derive the above-described sub-goal may be used to derive the final goal. In addition, in order to induce the i-th remark, the user 1's remark may be predicted and induced by using the similarity with the r-th remark (r is a natural number greater than i). In this case, it can be seen that the r-th remark replaces the sub-goal.

In an embodiment, the inequality sign of the above-mentioned formulas is indicated to include the same (=), but since it is difficult to derive a target when the similarity is repeated only when the same degree of similarity is repeated, the inequality sign may be deleted from the inequality sign.

In the present disclosure, an end goal, a sub-goal, a goal, and the like are exemplary, and the present disclosure may be interpreted as a dialogue method for inducing a specific goal.

The apparatus and method described above may be implemented as a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that can include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

The software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to a person skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. -magneto-optical media, and specially configured to store and carry out program instructions such as ROM, RAM, flash memory, etc.

Hardware devices are included. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

A computing device may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s), via wired and/or wireless communications. The remote computer(s) may be workstations, server computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, generally comprising the components described for the computing device. includes many or all of Logical connections include wired/wireless connections to local area networks (LANs) and/or larger networks, eg, wide area networks (WANs). Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

As described above, although the embodiments have been described with reference to the limited drawings, those skilled in the art may apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

1: User 10: Conversation System
100: user interface 200: natural language processor
300: processor 400: database
500: artificial intelligence module

Claims (10)

a user interface configured to receive a utterance from a user of the conversation system and to provide the utterance of the conversation system to the user;
a natural language processor configured to process the user's speech and convert it into user speech data that the dialog system can interpret;
a database configured to store at least one conversation domain, the conversation domain including system speech data;
a dialog system comprising an artificial intelligence module configured to generate a utterance of the dialog system with reference to the natural language processor and the database in response to receiving a utterance from the user;
the dialog system induces the user to speak a specific goal,
The artificial intelligence module induces the user to speak a plurality of sub-goals before inducing the user to speak the specific target,
and the artificial intelligence module is configured to set a degree of similarity between a sub-goal relatively close to the specific target and the specific target to be greater than or equal to a similarity between a sub-target relatively distant to the specific target and the specific target. According to claim 1,
The artificial intelligence module generates and provides the i-th utterance of the dialog system in response to the i-1 th utterance of the user,
A degree of similarity between each of one or more utterances of the user's utterances before the i-th utterance and the specific target, and a degree of similarity between the i-th utterance that the user is expected to make in response to the i-th utterance of the conversation system and the specific target and generate an i-th utterance of the conversation system based on 3. The method of claim 2,
The artificial intelligence module is
A degree of similarity between the user's i-th remark, which is expected to be spoken by the user in response to the i-th remark of the dialog system, and the specific target, and a degree of similarity between each of the user's k-th remarks to the i-1 th remarks and the specific target configured to generate the i-th utterance such that an average value of the sum of the user's l-th utterances to i-1 th utterances is equal to or greater than an average value of the similarity of the specific target to each of the user's l-th utterances - wherein k and l are less than i-1 and 1 Greater than natural numbers -, dialogue system. 3. The method of claim 2,
and the conversation system is configured to count the conversation i-1, and after counting the i-1 time, generate and provide the utterance i in response to the utterance i-1 of the user. According to claim 1,
the database is configured to store a plurality of conversation domains, each of the plurality of conversation domains being a conversation domain configured to derive a different specific goal,
and the dialog system is configured to select any one of the plurality of dialog domains. In a conversation method of a conversation system that induces a user to speak a specific goal,
The dialogue system is
a user interface configured to receive a utterance from a user of the conversation system and to provide the utterance of the conversation system to the user;
a natural language processor configured to process the user's speech and convert it into user speech data that the dialog system can interpret;
a database configured to store at least one conversation domain, the conversation domain including system speech data;
an artificial intelligence module configured to generate a utterance of the dialog system with reference to the natural language processor and the database in response to receiving a utterance from the user, the conversation method comprising:
before inducing the user to speak the specific target, inducing the user to speak a first sub-goal;
inducing the user to speak a second sub-goal after inducing the user to speak the first sub-goal and before inducing the user to speak the specific target; and
Inducing the specific target to be spoken,
a degree of similarity between the second sub-goal and the specific target is equal to or greater than the degree of similarity between the first sub-goal and the specific target; 7. The method of claim 6,
The artificial intelligence module further comprises the step of generating and providing the i-th utterance of the dialog system in response to the i-1 th utterance of the user,
The step of generating and providing the i-th utterance of the dialogue system includes:
A degree of similarity between each of one or more utterances of the user's utterances before the i-th utterance and the specific target, and a degree of similarity between the i-th utterance that the user is expected to make in response to the i-th utterance of the conversation system and the specific target and generating an i-th utterance of the conversation system based on 8. The method of claim 7,
The step of generating and providing the i-th utterance of the dialogue system includes:
A degree of similarity between the user's i-th remark, which is expected to be spoken by the user in response to the i-th remark of the dialog system, and the specific target, and a degree of similarity between each of the user's k-th remarks to the i-1 th remarks and the specific target calculating a first average value that is an average value of the sum of ;
calculating a second average value that is an average value of the similarity between each of the i-1 th utterances from the 1 th utterance of the user and the specific target; and
generating the i-th utterance such that the first average value is equal to or greater than the second average value, wherein k and l are natural numbers less than i-1 and greater than 1; 8. The method of claim 7,
Conversation, the dialog system further comprising: counting the conversation i-1 times; and generating the utterance i times in response to the user's utterance i-1 times after counting the i-1 times. Way. 7. The method of claim 6,
the database is configured to store a plurality of conversation domains, each of the plurality of conversation domains being a conversation domain configured to derive a different specific goal,
The conversation method further comprises the step of selecting any one of the plurality of conversation domains before the step of inducing the first sub-goal to speak.

Images