KR20200094311A - Method for providing chatbot service, apparatus and system thereof - Google Patents

Abstract

In developing a plurality of chatbots applied to a plurality of service domains, provided are a chatbot service providing method, apparatus, and system capable of reducing development costs and improving response performance of the chatbot. According to some embodiments of the present disclosure, a chatbot service providing method may comprise the steps of: determining a target service domain to which a target chatbot is applied from among a plurality of hierarchically defined service domains; obtaining a first conversation model for a higher service domain of the target service domain; inheriting the first conversation model and generating a second conversation model of the target service domain; and providing a chatbot service for the target service domain through the target chatbot using the second conversation model.

Description

METHOD FOR PROVIDING CHATBOT SERVICE, APPARATUS AND SYSTEM THEREOF}

The present disclosure relates to a method, apparatus and system for providing a chatbot service. More specifically, in providing a chatbot service in a plurality of service domains, a method for reducing chatbot development and management costs, and improving a chatbot's responsiveness, and a device and system supporting the method.

Chatbot (chatbot) is a combination of chat and robot, and refers to software designed to perform conversations with humans through text or voice and to perform specific tasks through dialogue. Chatbots are called by various names such as chatterbots, talkbots, intelligent agents, and interactive agents.

Recently, chatbots have been introduced into various reservation and consultation services, replacing some of the agents' work. In particular, the chatbot for counseling provides a 24-hour reservation and counseling service using a messenger platform that is easily accessible even on mobile devices. Through this, human costs incurred in the counseling process are reduced, and the simplicity and effectiveness of service delivery are being improved.

However, in the case of introducing chatbots for reservation and consultation services to various service domains, since the chatbots must be developed for each service domain, a development cost increases rapidly. For example, when developing a chatbot that provides consultation and reservation services for multiple hotels, a separate chatbot must be developed for each hotel, so the development cost of the chatbot increases rapidly. In addition, since iterative development is performed on a common part, waste of development cost is caused.

The above development cost problem is the biggest barrier to entry in the introduction of chatbots in various service domains. Therefore, in developing a chatbot that is used in multiple service domains, a method for reducing development cost and improving development efficiency is required.

Korean Open Patent No. 10-2018-0089242 (published August 8, 2018)

The technical problem to be solved through some embodiments of the present disclosure is a method for reducing development cost of a chatbot, a device and a system supporting the method, in developing a plurality of chatbots to be used for a plurality of service domains Is to provide.

Another technical problem to be solved through some embodiments of the present disclosure, in providing a chatbot service in a plurality of service domains, is a method for improving the response performance of a chatbot, and providing an apparatus and system supporting the method will be.

Another technical problem to be solved through some embodiments of the present disclosure, in providing a chatbot service in a plurality of service domains, a method for reducing the human cost of updating a chatbot conversation model and supporting the method To provide devices and systems.

Another technical problem to be solved through some embodiments of the present disclosure, in providing a chatbot service in a plurality of service domains, a method for reducing the human cost of managing a chatbot conversation model and supporting the method To provide devices and systems.

The technical problems of the present disclosure are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above technical problem, a method for providing a chatbot service according to some embodiments of the present disclosure includes a target in which a target chatbot is applied among a plurality of hierarchically defined service domains in a method for providing a chatbot service performed by a computing device. Determining a service domain, obtaining a first conversation model for an upper service domain of the target service domain, inheriting the first conversation model, generating a second conversation model of the target service domain, and the second And providing a chatbot service for the target service domain through the target chatbot using a conversation model.

In some embodiments, providing the chatbot service includes receiving a query from a user terminal, generating a response to the query using the second conversation model, and sending the generated response to the user terminal. And providing steps.

In some embodiments, providing the chatbot service includes receiving a query from a user terminal, determining whether a response to the query is possible based on the second conversation model, and determining that the response is not possible In response, selecting a reference service domain from among the plurality of service domains based on the similarity between service domains and providing a response generated using the third conversation model associated with the reference service domain to the user terminal. It can contain.

A method for providing a chatbot service according to some other embodiments of the present disclosure for solving the above technical problem is a method of providing a chatbot service using a plurality of conversation models utilized in a plurality of service domains on a computing device. Receiving a query from a user terminal through a chatbot, determining whether a response to the query is possible based on a conversation model of a target service domain to which the target chatbot is applied, and in response to a determination that the response is not possible, a service The method may include selecting a reference conversation model from among the plurality of conversation models based on similarity between domains, and providing a response generated using the reference conversation model through the target chatbot.

In some embodiments, the selecting of the conversation model for reference may include selecting a conversation model of a service domain that satisfies a criterion similarity allowed for the received query among the plurality of service domains as the reference conversation model. It can contain.

In some embodiments, selecting the conversation model for reference may include selecting a plurality of candidate service domains among a plurality of service domains based on the similarity, and using a plurality of conversation models of each of the plurality of candidate service domains. It may include generating a response, evaluating the quality of each of the plurality of responses, and selecting the dialogue model for reference based on the evaluation result.

In some embodiments, the method may further include obtaining feedback on the generated response from the user terminal and adjusting the similarity of a service domain associated with the reference conversation model based on the feedback.

In some embodiments, the similarity may be calculated based on the similarity between the first specification information of the target service domain and the second specification information of another service domain.

A method of providing a chatbot service according to another exemplary embodiment of the present disclosure for solving the above technical problem is provided in a method of providing a chatbot service using a plurality of chatbots utilized in a plurality of service domains on a computing device, wherein Obtaining query data including an unanswered query from a first chatbot utilized in a first service domain among a plurality of service domains, receiving a recommendation response to the query data, and a service different from the first service domain Recommending at least a part of the plurality of service domains as an update target based on the similarity between domains, and updating a conversation model associated with the recommended service domain using update data including the query data and the recommendation response It may include steps.

In some embodiments, the query data may further include a query that has received negative user feedback.

In some embodiments, the step of recommending as an update target includes recommending the at least a portion based on the performance of a conversation model associated with each service domain, wherein the performance of the conversation model is associated with each conversation model. It can be determined based on the number of scenarios.

In some embodiments, recommending the update target includes recommending the at least a portion based on the performance of a conversation model associated with each service domain, wherein the performance of the conversation model is the number of unanswered queries. And based on user feedback on the answered query.

In some embodiments, the plurality of service domains are defined in a hierarchical structure, and the step of recommending the update target may include, in a first layer where the first service domain is located, similarity between the service domains of a service domain having a reference value or higher. In response to determining that the count is counted and the count is greater than or equal to a threshold, recommending a second service domain located in an upper layer of the first layer as the update target, and the updating step includes: Updating a first conversation model associated with the second service domain using the update data and batch updating a conversation model associated with a lower service domain of the second service domain based on the updated first conversation model It may include the steps.

A computing device according to some embodiments of the present disclosure for solving the above-described technical problem, a communication interface communicating with a user terminal and a target chatbot, a memory including one or more instructions, and executing the one or more instructions By determining a target service domain to which the target chatbot is applied among a plurality of hierarchically defined service domains, obtaining a first conversation model for an upper service domain of the target service domain, and inheriting the first conversation model And generating a second conversation model of the target service domain and providing a chatbot service for the target service domain through the target chatbot using the second conversation model.

A computing device according to some other embodiments of the present disclosure for solving the above-described technical problem includes a communication interface communicating with a user terminal and a target chatbot, a memory including one or more instructions, and the one or more instructions. By executing, it receives a query from a user terminal through the target chatbot, determines whether it is possible to respond to the query based on the conversation model of the target service domain to which the target chatbot is applied, and responds to the determination that the response is not possible A processor for selecting a reference conversation model from among a plurality of conversation models associated with different service domains based on similarity between service domains and providing a response generated using the reference conversation model through the target chatbot Can.

Computing device according to another some embodiments of the present disclosure for solving the above-described technical problems, a communication interface to communicate with one or more chatbot server providing a chatbot service using a plurality of chatbots utilized in a plurality of service domains, By executing a memory including one or more instructions and the one or more instructions, query data including an unanswered query is obtained from a first chatbot utilized in a first service domain among the plurality of service domains, Receiving a recommendation response for the query data, recommending at least a portion of the plurality of service domains as an update target based on the similarity between the first service domain and another service domain, and including the query data and the recommendation response And a processor for updating a conversation model associated with the recommended service domain using update data.

Computing program in accordance with some embodiments of the present disclosure for solving the above-described technical problem, determining a target service domain to which a target chatbot is applied among a plurality of hierarchically defined service domains in combination with a computing device, the target Obtaining a first conversation model for an upper service domain of a service domain, inheriting the first conversation model, generating a second conversation model of the target service domain, and through the target chatbot using the second conversation model It may be stored in a computer-readable recording medium to execute the step of providing a chatbot service for the target service domain.

Computing program according to some other embodiments of the present disclosure for solving the above-described technical problem is coupled to a computing device, receiving a query from a user terminal through a target chatbot, the conversation of the target service domain to which the target chatbot is applied Determining whether a response to the query is possible based on a model, and in response to a determination that the response is not possible, selecting a reference conversation model from among a plurality of conversation models associated with different service domains based on similarities between service domains And it may be stored in a computer-readable recording medium to execute the step of providing a response generated using the reference chat model and the target chatbot.

In yet another exemplary embodiment of the present disclosure for solving the above-described technical problem, a computing program includes an unanswered query from a first chatbot that is combined with a computing device and utilized in a first service domain among the plurality of service domains. Obtaining query data to be received, receiving a recommendation response to the query data, recommending at least a portion of a plurality of service domains as an update target based on the similarity between the first service domain and another service domain, and It may be stored in a computer-readable recording medium to execute a step of updating a conversation model associated with the recommended service domain using update data including query data and the recommendation response.

1 is an exemplary configuration diagram illustrating a chatbot service providing system and a network environment according to some embodiments of the present disclosure.
2 is an exemplary view for explaining a chatbot service that can be referred to in various embodiments of the present disclosure.
3 is an exemplary diagram illustrating a hierarchical structure of a service domain according to some embodiments of the present disclosure.
4 and 5 are exemplary views for explaining a conversation model that can be referred to in various embodiments of the present disclosure.
6 is an exemplary flowchart illustrating a method for providing a chatbot service according to some embodiments of the present disclosure.
FIG. 7 is an exemplary diagram for explaining the dialogue model generation step S200 illustrated in FIG. 6.
FIG. 8 is an exemplary detailed flow chart showing the chatbot service providing step S300 illustrated in FIG. 6.
9 is an exemplary diagram for illustrating a concept of similarity between service domains according to some embodiments of the present disclosure.
10 and 11 are exemplary diagrams for describing a method for searching a dialogue model for reference based on similarity according to some embodiments of the present disclosure.
12 and 13 are exemplary views for describing a method of calculating similarity between service domains according to some embodiments of the present disclosure.
14 is an exemplary diagram for describing a method for adjusting similarity based on user feedback according to some embodiments of the present disclosure.
15 and 16 are exemplary flowcharts illustrating a method for updating a conversation model according to various embodiments of the present disclosure.
FIG. 17 is an exemplary diagram for explaining the update target recommendation step S530 illustrated in FIG. 16.
18 shows an example computing device capable of implementing devices in accordance with various embodiments of the present disclosure.

Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Advantages and features of the present disclosure, and a method of achieving them will be apparent with reference to embodiments described below in detail together with the accompanying drawings. However, the technical spirit of the present disclosure is not limited to the following embodiments, but may be implemented in various different forms, and only the present embodiments allow the present disclosure to be complete, and common knowledge in the technical field to which the present disclosure pertains. It is provided to fully inform the holder of the scope of the present disclosure, and the technical spirit of the present disclosure is only defined by the scope of the claims.

It should be noted that in adding reference numerals to the components of each drawing, the same components have the same reference numerals as possible, even if they are displayed on different drawings. In addition, in describing the present disclosure, when it is determined that detailed descriptions of related well-known structures or functions may obscure the subject matter of the present disclosure, detailed descriptions thereof will be omitted.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used in a sense that can be commonly understood by those skilled in the art to which this disclosure belongs. In addition, terms defined in the commonly used dictionary are not ideally or excessively interpreted unless specifically defined. The terminology used herein is for describing the embodiments and is not intended to limit the present disclosure. In the present specification, the singular form also includes the plural form unless otherwise specified in the phrase.

Further, in describing the components of the present disclosure, terms such as first, second, A, B, (a), and (b) may be used. These terms are only for distinguishing the component from other components, and the nature, order, or order of the component is not limited by the term. When a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected to or connected to the other component, but another component between each component It should be understood that elements may be "connected", "coupled" or "connected".

As used herein, "comprises" and/or "comprising" refers to the elements, steps, operations and/or elements mentioned above, the presence of one or more other components, steps, operations and/or elements. Or do not exclude additions.

Prior to the description of the present specification, some terms used in the specification will be clarified.

In this specification, a service domain is an abstract term that refers to an area that provides a specific service. The service domain may be defined in various ways, and multiple sub-domains may be included in one service domain. For example, a first sub-domain associated with hotel A, a second sub-domain associated with hotel B, and the like may be included in the hotel service domain. The service domain may be used interchangeably with terms such as a business domain in the technical field, and for an example of the service domain, refer to FIG. 3.

In this specification, the chatbot service (chatbot service) literally means a service provided through a chatbot. For example, the chatbot service may be a consultation service or a reservation service. However, the technical scope of the present disclosure is not limited to the examples listed above.

In this specification, a conversation model (dialog model or conversation model) is a model used to generate a response to a query. The conversation model may include various types of models, such as rule-based, scenario-based, and deep-learning-based. However, the technical scope of the present disclosure is not limited to the examples listed above. The conversation model may be used interchangeably with terms such as a query/response model, a conversation engine, and a chatbot engine in the art.

In this specification, an instruction is a series of instructions grouped based on a function and refers to a component of a computer program and executed by a processor.

Hereinafter, some embodiments of the present disclosure will be described in detail according to the accompanying drawings.

1 is an exemplary configuration diagram illustrating a chatbot service providing system and a network environment according to some embodiments of the present disclosure.

As illustrated in FIG. 1, the chatbot service providing system may include one or more chatbot servers 20-1 to 20-n and a management server 30. However, this is only a preferred embodiment for achieving the object of the present disclosure, and of course, some components may be added or deleted as necessary. In addition, it is noted that each component of the chatbot service providing system illustrated in FIG. 1 is functionally divided functional elements, and a plurality of components may be implemented in an integrated form in a physical environment. For example, at least some of the one or more chatbot servers 20-1 to 20-n and the management server 30 may be implemented in the form of different logic in one physical computing device.

Further, in the actual physical environment, each of the components may be implemented in a form of being divided into a plurality of detailed functional elements. For example, the first function of the chatbot server (e.g. 20-1) may be implemented in the first computing device, and the second function may be implemented in the second computing device. In the same context, multiple functions of the management server 30 may also be implemented in multiple computing devices.

Hereinafter, for convenience of description, when referring to one or more chatbot servers 20-1 to 20-n generically or to any chatbot server (eg 20-1 or 20-2 or 20-n), reference number "20" To use. In addition, when referring to one or more user terminals 10-1 to 10-n or an arbitrary user terminal (e.g. 10-1 or 10-2 or 10-n), reference number "10" is used. Hereinafter, each component of the chatbot service providing system will be described.

In the chatbot service providing system, the chatbot server 20 is a computing device that provides a chatbot service. For example, the chatbot server 20 may receive a query from the user terminal 10 and provide a response to the query through the chatbot. The chatbot service may be preferably provided through a messenger platform to maximize user accessibility, but the technical scope of the present disclosure is not limited thereto.

The computing device may be a laptop, a desktop, a laptop, etc., but is not limited thereto, and may include all types of devices equipped with computing and communication functions. 18 for some examples of the computing device.

The plurality of chatbot servers 20 may provide chatbot services in a plurality of service domains. For example, as illustrated in FIG. 2, the plurality of chatbot servers 20 make reservations and consultations in service domains associated with multiple hotels 41-1 to 41-n through chatbots 43-1 to 43-n. Can provide services. For example, the first chatbot server 20-1 may operate to provide a reservation and consultation service (see 45) for the A hotel 41-1 through the first chatbot 43-1.

According to various embodiments of the present disclosure, a service domain may be defined in a hierarchical structure in order to minimize development costs for a plurality of chatbots utilized in a plurality of service domains. In order to provide convenience of understanding, a hierarchical structure of a service domain will be described in detail with reference to an example illustrated in FIG. 3. In the tree-like hierarchical structure shown in FIG. 3, each node of the tree corresponds to a service domain.

As shown in FIG. 3, the hierarchical structure 50 of the service domain includes a service domain of a relatively generalized upper layer (hereinafter abbreviated as “higher service domain”) and a service domain of a relatively concrete lower layer (hereinafter referred to as “service domain”). , Abbreviated as "sub-service domain"). In addition, a conversation model to be used for a chatbot for each service domain may be generated.

As the service domains are hierarchically structured, the conversation model of the parent service domain (hereinafter abbreviated as "parent conversation model") and the conversation model of the lower service domain (hereinafter abbreviated as "child conversation model") are inherited. (Ie, hierarchical relationship). In other words, the child conversation model includes the parent conversation model (ie, parent or ancestor node) associated with it, and the parent conversation model is created to include the common part of the child conversation model (ie, child node or descendant node) associated with it. Can be. For example, the conversation model of the B hotel node 59 may include the conversation model of the Seoul node 57, which is an upper service domain.

Due to the above characteristics, when the conversation model of the target service domain is generated, the conversation model of the upper service domain is inherited as it is, and only a portion specialized for the target service domain is added, so that development cost for chatbot development can be greatly reduced. have.

In the hierarchical structure of the service domain, each layer can be flexibly defined, which may vary depending on the embodiment. For example, as illustrated in FIG. 3, the first layer 51 is defined as a service domain of a higher concept, and the lower layers 53 and 55 define a service domain of the higher concept by region (eg Seoul area, Jeju). Region), or by company (eg A Hotel, B Hotel, etc.). In this case, the conversation model of the first layer 53 is generated to provide a response function to a query common to the corresponding service domain regardless of region, and the conversation model of the second layer 53 is specific to each service domain. It may be created to provide a response function for a common query in a region. In addition, the conversation model of the third layer 55 may be generated to provide a response function for a query specialized for a specific company in a specific region. However, the technical scope of the present disclosure is not limited thereto.

The conversation model may be implemented in various forms, which may vary depending on the embodiment.

In some embodiments, the conversation model may be a rule-based or scenario-based model. That is, the conversation model may be a model including a plurality of rules or scenarios and generating a response based on rules or scenarios matching the input query. The rule may be defined to include a query and a response thereto, but the technical scope of the present disclosure is not limited thereto. The scenario may be defined to further include a dialog flow along with a query/response, but the technical scope of the present disclosure is not limited thereto. In the present embodiment, the upper conversation model includes a common scenario or common rules of the lower conversation model, and the lower conversation model may inherit the upper hierarchical conversation model in a form of duplicating the common scenario or the common rules.

In some embodiments, the above-described rules or scenarios may be defined by intent. For example, as shown in FIG. 4, the conversation model may include one or more scenarios 63-1 to 63-n associated with the first intention 61 (“room reservation”), and one or more scenarios associated with the second intention. Can. In such a case, the conversation model may generate a response based on a rule or scenario matching the query among candidate rules or candidate scenarios matching the intention included in the query.

In some embodiments, the above-described rules or scenarios are defined in the form of a dialog frame having a plurality of slots, and a chat-bot service may be provided in a slot-filling manner. For example, if the information required for room reservation is defined as a slot, such as the conversation frame 70 illustrated in FIG. 5, the conversation model may generate a response sentence for a query by filling information of each slot.

In some other embodiments, the conversation model may be a model based on machine-learning techniques such as deep learning. That is, the conversation model may be a model constructed by using a set of conversation pairs consisting of query/response as training data. In this case, the lower conversation model inherits the upper conversation model through transfer learning (i.e., inherits the weight of the trained model) or the upper conversation model by learning the training dataset of the upper layer. Will be able to receive

A detailed description of how the chatbot server 20 provides the chatbot service will be described later with reference to the drawings of FIG. 6 and below. It will be described again with reference to FIG. 1.

In the chatbot service providing system, the management server 30 is a computing device that performs overall management and monitoring for the chatbot server 20. The computing device may be a laptop, a desktop, a laptop, etc., but is not limited thereto, and may include all types of devices equipped with computing and communication functions. However, in the case of performing management for a plurality of chatbot servers 20, the management server 30 may be preferably implemented as a high-performance server-class computing device. 18 for some examples of the computing device.

In some embodiments, the management server 30 may receive historical data from the chatbot server 20, which may include data related to unanswered queries and data related to queries/responses that received negative feedback. Can. The management server 30 may update the conversation model used by the chatbot server 20 based on the history data. A detailed description of this embodiment will be described later with reference to FIGS. 16 and 17.

In some other embodiments, the chat model is updated by the chatbot server 20 itself, and the management server 30 can receive the update result. In addition, the management server 30 may update the conversation model of another service domain based on the update result. A detailed description of this embodiment will be described later with reference to FIG. 15.

The user terminal 10 is a user terminal using a chatbot service. For example, the user may use the chatbot service in the form of sending a query through the terminal 10 and receiving a response thereto. At this time, the interface with the chatbot can be implemented in various forms such as messenger-based and web-based. The user terminal 10 may be implemented with any device

As shown in FIG. 1, the user terminal 10 and the chatbot service providing system can communicate through a network. In some embodiments, chatbot server 20 and management server 30 may also communicate over a network. Here, the network is a wired/wireless network of any kind, such as a local area network (LAN), a wide area network (WAN), a mobile radio communication network, a Wibro (Wireless Broadband Internet), and the like. Can be implemented.

So far, a chatbot service providing system and a network environment according to some embodiments of the present disclosure have been described with reference to FIGS. 1 to 5. Hereinafter, a method for providing a chatbot service according to various embodiments of the present disclosure will be described in detail. In order to provide the convenience of understanding, the description of the above method will be continued assuming the environment shown in FIG. 1, but those skilled in the art can clearly understand that the environment in which the chatbot service is provided can be modified in any way. There will be.

Each step of the methods described below can be performed by a computing device. In other words, each step of the methods may be implemented with one or more instructions executed by a processor of a computing device. All steps included in the methods may be performed by one physical computing device, but the first steps of the method are performed by the first computing device, and the second steps of the method are performed by the second computing device. It may be performed by. Hereinafter, unless otherwise stated, it is assumed that each step of the method is performed by the chatbot server 20 to continue the description. However, for convenience of description, description of the operation subject of each step included in the method may be omitted. In addition, it is needless to say that the methods to be described later may change the execution order of each operation within a range in which the execution order can be logically changed as necessary.

6 is an exemplary flowchart illustrating a method for providing a chatbot service according to some embodiments of the present disclosure. In particular, FIG. 6 illustrates a process of developing a chatbot of a target service domain (e.g. B hotel's service domain) and providing a chatbot service through the developed chatbot. However, this is only a preferred embodiment for achieving the object of the present disclosure, and of course, some steps may be added or deleted as necessary.

As illustrated in FIG. 6, the method for providing a chatbot service starts in step S100 of determining a target service domain to be applied to a target chatbot on a hierarchical structure of service domains. Referring to the example illustrated in FIG. 7, it is assumed that a chatbot 81 for a B hotel located in Seoul is developed. In this case, the service domain 83 associated with the B hotel on the hierarchical structure 80 may be determined as the target service domain. If the target service domain does not exist in the hierarchical structure, the administrator may add the target service domain to an appropriate location in the hierarchical structure by considering the relationship between the target service domain and other service domains.

In step S200, a conversation model (hereinafter, abbreviated as "target model") to be used for the target service domain is generated by inheriting the upper conversation model for the target service domain. For example, referring back to FIG. 7, a target model may be generated by inheriting the conversation model of the upper service domain 85 for the target chatbot 81.

In this step S200, if data (e.g. conversation pairs, scenarios, rules associated with the e.g. B hotel) specific to the target service domain are prepared, the data may be further reflected in the target model.

In some embodiments, a tool such as a chatbot builder may be provided to provide convenience of chatbot development. In this case, the aforementioned steps S100 and S200 may be performed together with the computing device equipped with the chatbot builder. Further, in order to unify management of the conversation model, if the conversation model is managed by the management server 30, the above-described steps S100 and S200 may be performed in conjunction with the management server 30. In this case, the management server 30 may operate to distribute the target model to the chatbot server 20 on which the target chatbot operates.

In step S300, a chatbot service for a target service domain is provided through a target chatbot using a target model. For example, the chatbot server 20 receives a query for the B hotel through the target chatbot, generates a response to the received query using a target model, and provides the generated response through the target chatbot to the B hotel It can provide a chatbot service for.

In some embodiments, various scenarios included in the dialogue model may be defined based on intention (see FIG. 4). In such a case, the chatbot server 20 extracts the intention included in the received query, determines a scenario matching the intention through the target model, and generates and provides a response to the query based on the determined scenario Can. The extraction of the intention may be performed through a model based on machine learning, but the technical scope of the present disclosure is not limited thereto.

A more detailed description of this step S300 will be described later with reference to FIG. 8.

So far, a method of providing a chatbot service according to some embodiments of the present disclosure has been described with reference to FIGS. 6 and 7. According to the above-described method, by recycling the upper dialogue model composed of the common part of the lower dialogue model, an effect of reducing development cost for chatbot development can be achieved.

Hereinafter, a specific process in which the chatbot service is provided in step S300 will be described with reference to FIGS. 8 to 14.

8 is an exemplary detailed flow diagram illustrating a chatbot service providing step S300 according to some embodiments of the present disclosure.

As illustrated in FIG. 8, the chatbot service providing step S300 starts at step S310 in which a query is received through the target chatbot. For example, a query may be received from the user terminal 10 through the target chatbot.

In step S320, it is determined whether a response to the received query is possible based on the target model. If the answerable decision is made, step S330 is performed, and in the opposite case, step S350 is performed.

In steps S330 and S340, in response to the determination that the response is possible, a response is generated using the target model, and the generated response is provided through the target chatbot. For example, the response may be provided to the user terminal 10 through the target chatbot.

In step S350, in response to the non-response determination, a reference conversation model is searched. The reference conversation model means a conversation model of another service domain capable of performing a response to a received query. The details of the step S350 will be described in detail later with reference to FIGS. 9 to 14.

In steps S360 and S370, if the search is successful (that is, if a reference conversation model is found), a response to a query received through the reference conversation model is generated. The generated response is provided through the target chatbot (S340).

In steps S360 to S390, if the search fails (i.e., if the conversation model for reference is not found), an apology message is provided through the target chatbot. In addition, unanswered queries are reported to the management server 30. Of course, in some other embodiments, the answered query and user feedback information for the response may also be transmitted to the management server.

So far, a detailed process in which the chatbot service is provided has been described with reference to FIGS. 7 and 8. According to the above, even for a query that cannot be answered through the target model, an appropriate response can be performed through searching the reference conversation model. Accordingly, the response performance of the chatbot is improved, and the satisfaction of the user receiving the chatbot service can be improved.

Hereinafter, a method of searching a reference conversation model will be described with reference to FIGS. 9 to 11.

According to various embodiments of the present disclosure, a conversation model for reference may be searched based on similarity between service domains. For convenience of understanding, this embodiment will be described in detail with reference to FIG. 9. In the hierarchical structure of the service domain shown in the figure of FIG. 9 or less, numbers shown adjacent to the dotted arrows indicate "similarity between service domains."

9, the similarity between the target service domain 91 and other service domains 93 to 97 on the hierarchical structure 90 may be preset. The similarity may be set only for some service domains or may be set for all. The similarity refers to the similarity according to the characteristics of the service domain, and may be set by an administrator who is familiar with the characteristics of each domain, or may be automatically set based on specification information of each domain. The method for automatically calculating the similarity will be described later.

In some embodiments, a search may be sequentially performed for service domains having a similarity between the service domains above a threshold. That is, the search may be sequentially performed from a domain having high similarity using the threshold as a search termination condition. For example, when the threshold is 65%, the A hotel domain 93 having the highest similarity is searched, and then a search for the B hotel domain 97 can be made. Here, the fact that the search is performed includes determining whether it is possible to respond to a query received through a conversation model of another service domain (e.g. 93, 97). If a responsive conversational model (i.e., a reference conversational model) is found, the search process may end immediately.

In some examples, additional termination conditions may be set to minimize search delay. For example, when it is desired to limit the number of searches, the termination condition (e.g., after one search and termination) according to the number of searches may be set as an additional termination condition. For another example, when you want to limit the search range, the termination condition according to the performance of the conversation model (e.g., the domain of the conversation model whose performance is below a threshold is excluded from the search) may be set as an additional termination condition. The performance of the conversation model includes the number of rules/scenarios (or the degree of learning) included in the conversation model, the number of queries answered, the number of unanswered queries, the number of queries that received positive feedback, and the number of queries that received negative feedback. It can be determined on the basis of. For example, as the number of scenarios included in the conversation model increases, the number of unanswered queries decreases, and the number of negative feedback decreases, the performance of the conversation model may be determined to be higher. However, the technical scope of the present disclosure is not limited thereto, and it is needless to say that the performance of the dialogue model may be determined by using other indicators.

In some examples, similarity between service domains may be adjusted based on the performance of the conversation model. For example, as the performance of the conversation model increases, the similarity may be adjusted to a high value. This is because as the performance of the conversation model increases, the probability of response to the received query will improve.

In some examples, a history of previously performed searches may be utilized to minimize search delay. The search history may include information such as a query not answered by the target model, a conversation model for reference to the query, and user feedback. According to the present embodiment, an effect of minimizing the search delay can be achieved by using the search history for the similar query.

In some other embodiments, a plurality of responses may be generated for a received query using all (or only a portion) of a conversation model of a service domain (or an entire service domain) having similarity between service domains having a threshold level or higher. Also, a conversation model for reference may be determined based on the quality evaluation results for the plurality of responses. For example, as shown in FIG. 10, a plurality of responses (a, b, c) are generated from the conversation models of all service domains 103 to 107 having similarity with the target service domain 101 on the hierarchical structure 100. Can be obtained. In addition, quality evaluation of a plurality of responses (a, b, c) is performed, and a response (b) having the highest quality may be provided as a response to the received query.

The specific manner of performing the quality evaluation on the response may vary depending on the embodiment. For example, the quality evaluation includes the length of the response (eg, the higher the quality), the number of nouns (or entities) included in the response (eg the higher the higher the quality), the association between the intention included in the query and the response (eg the higher the higher the quality), Among the words included in the response, it may be performed based on an indicator such as the number of words (eg, the higher the higher the quality) associated with the target service domain, but the technical scope of the present disclosure is not limited thereto.

In some examples, an evaluation score may be calculated according to the evaluation indexes listed above, and a final evaluation score may be calculated by reflecting the similarity set in each service domain in the evaluation score (e.g., using the similarity as a weight). In addition, the response with the highest final evaluation score may be provided through the target chatbot.

On the other hand, in some embodiments of the present disclosure, reference similarity may be set for each query (or scenario). The reference similarity refers to a minimum similarity between service domains allowed for each query, and is a value used to limit search ranges between service domains differently for each query. The present embodiment will be described in detail with reference to the table 110 shown in FIG. 11.

Referring to FIG. 11, query #1 has a reference similarity of "50". Therefore, when searching the reference conversation model for query #1, the search can be performed only for service domains having a similarity between service domains of at least 50% or higher. For another example, query #3 has a standard similarity of "90". Therefore, when searching the reference conversation model for query #3, the search can be performed only for service domains having a similarity between service domains of at least 90% or higher.

In some examples, the criterion similarity may be determined based on the strictness required for the response (ie, the extent to which an erroneous response affects the user). For example, in the case of queries closely related to money, incorrect responses can have a great impact on users. In this case, the standard similarity of the query can be set to a very high value.

So far, a method of searching a conversation model for reference has been described with reference to FIGS. 9 to 11. According to the above-described method, a model capable of performing the most accurate response based on the similarity between service domains may be determined as a reference conversation model. Accordingly, the response performance of the chatbot can be improved, and the satisfaction of customers who are provided with the chatbot service can be improved.

Hereinafter, a method of automatically calculating and adjusting the similarity between the above-described service domains will be described with reference to FIGS. 12 to 14. First, a method of automatically calculating similarity between service domains will be described with reference to FIGS. 12 and 13.

Since the similarity between the service domains is a key indicator for determining a reference conversation model, it is important to accurately calculate the value. This is because if the reference conversation model is incorrectly determined, an inappropriate response will be provided to the user. It is desirable that the similarity is set by an administrator who understands each service domain well, but as the number of service domains increases, the manual setting method becomes inefficient and increases the probability of an error.

In order to solve the above problems, in some embodiments of the present disclosure, the similarity between service domains may be automatically calculated based on the similarity between specification information of service domains. To provide convenience of understanding, the present embodiment will be described in detail with reference to FIG. 12. In particular, FIG. 12 shows an example of calculating similarity between domains associated with hotels A and B.

As illustrated in FIG. 12, when detailed specification information 120 for a service domain to which a chatbot is applied is secured, similarity between service domains may be calculated based on the specification information 120. In the case of hotel services, the specification information includes specification information (eg, the size and number of restaurants, size and number of meeting rooms), transportation-related specification information (eg, the size and number of shuttles), etc. Specification information about the service (eg, whether breakfast is provided, type of room service, etc.), and location of the hotel may be included. However, the specification information listed above is only for describing some embodiments of the present disclosure, and the technical scope of the present disclosure is not limited to the specification information listed above.

12, the similarity between the A hotel domain 123 and the B hotel domain 125 may be calculated according to the similarity between the specification information 121 of the A hotel and the B hotel. Any method may be used as a specific method of calculating the similarity between specification information.

In this embodiment, the reason for using the specification information as core data for similarity between service domains is that both query and response for a specific service domain are closely related to specification information of the specific service domain. For example, suppose that a similar size hotel in a similar area offers a chatbot service. Then, queries received through the chatbots of the two hotels are relatively similar to other sizes of hotels located in different regions, and the responses are also similar. Accordingly, when the similarity between service domains is calculated using the specification information, the similarity can be accurately calculated, and accordingly, the search time for the reference conversation model can also be reduced.

In some embodiments, similarities between service domains may be subdivided according to user intention. For example, the similarity between the first service domains may be calculated for the first user intention, and the similarity between the second service domains may be calculated for the second user intention. Further, the similarity between the first service domains may be calculated based on specification information associated with the first user intention. In order to provide more convenience, the present embodiment will be described in detail with reference to FIG. 13. 13 illustrates calculating similarity between service domains for user intention ("room inquiry").

As illustrated in FIG. 13, when calculating similarity between service domains for a room inquiry, specification information 130 for the room may be used. That is, the similarity between the A hotel domain 133 and the B hotel domain 135 for room inquiry may be calculated based on the room specification information 131 of the A hotel and the B hotel. The reason for using only the specification information related to the user's intention is to solve a problem in which the accuracy of similarity calculation is deteriorated by reflecting the specification information irrelevant to the user's intention.

In this embodiment, the intention included in the received query is extracted, and a search for a conversation model for reference can be made based on the similarity between service domains corresponding to the extracted intention. Accordingly, a model capable of more accurately responding to the received query may be determined as a reference conversation model, and the time required to search the model may also be reduced.

Meanwhile, since the similarity between service domains calculated based on the specification information corresponds to a kind of predicted value, an actual similarity and error may exist. In addition, due to the above error, when a real chatbot service is provided, a problem in which a conversation model for reference is incorrectly selected may occur. Accordingly, according to some embodiments of the present disclosure, a process of gradually reducing the error through a similarity adjustment based on user feedback is performed. Hereinafter, this embodiment will be described with reference to FIG. 14.

14 shows a process in which the chatbot of the target service domain 141 provides a response using the conversation model of the A hotel domain 143, and adjusts the similarity of the A hotel domain 143 based on feedback on the response. Is illustrated.

As illustrated in FIG. 14, when feedback is obtained from the user terminal (e.g. 10 in FIG. 1 ), the similarity may be adjusted based on the feedback. For example, similarity may increase when positive feedback is obtained, and similarity may decrease when negative feedback is obtained. The above feedback can be obtained by analyzing the sentences (e.g. thank you, etc.) entered after the response was provided. For example, feedback may be obtained through opinion mining on the input sentence. However, the technical scope of the present disclosure is not limited to this, and a method of obtaining the user feedback may vary.

In some examples, the similarity may be adjusted differentially according to the degree of polarity (ie, positive or negative) included in the feedback.

In some embodiments, if user feedback is not received within a designated time, feedback for the received query may be predicted. For example, if a query of the same intention as the received query is repeated, the feedback may be predicted as negative. For another example, if an intention different from the received query is included in the next query, the feedback may be predicted as positive. The inclusion of other intentions in the next query could be interpreted as providing an appropriate response to the previous query. For another example, if there is no additional query in the conversation scenario, there may be additional queries, and the feedback may be predicted as negative. In addition, user feedback may be predicted through the opinion mining result of the sentence received after the response.

In some examples, the adjustment width of similarity by predictive feedback may be smaller than the adjustment width by actual feedback. There is also a possibility that feedback may be mispredicted.

In some examples, when a response to be provided to a user is determined based on a quality evaluation result of a plurality of candidate responses (see FIG. 10 ), similarity may be adjusted to reflect the quality evaluation result. For example, the higher the quality evaluation score, the greater the degree of similarity increase.

So far, a method of calculating and adjusting similarity between service domains has been described with reference to FIGS. 12 to 14. According to the above-described method, the similarity between service domains is automatically calculated, so that the convenience of the administrator can be greatly improved. In addition, an error included in the similarity may be gradually corrected through an adjustment process based on feedback. Accordingly, the conversation model for reference can be accurately selected, and the response accuracy of the chatbot can be improved.

Hereinafter, a method for updating a conversation model according to various embodiments of the present disclosure will be described with reference to FIGS. 15 to 17.

15 is an exemplary flow diagram illustrating a method for updating a conversation model in accordance with some embodiments of the present disclosure. In particular, FIG. 15 illustrates a method of updating the target model by itself in the chatbot server 20.

As illustrated in FIG. 15, the method for updating the conversation model starts at step S400 in which a response to a query received through a target chatbot is provided. At this time, the response means a response generated using a reference conversation model.

In step S410, it is determined whether user feedback for the response is received. For example, if a sentence associated with feedback is not received through the target chatbot within a designated time, it may be determined that the feedback has not been received. If feedback is received, step S430 is performed. In the opposite case, step S420 is performed.

In step S420, feedback for the response is predicted. For details of the method for predicting feedback in step S420, the above-described embodiment will be referred to.

In step S430, it is determined whether the received (or predicted) feedback is positive. If it is determined positive, step S450 is performed. In the opposite case, step S440 is performed.

In step S440, historical data including the received query and response and feedback information is reported to the management server 30. Of course, even when the feedback is positive, it is a matter of course that the history data as described above may be reported to the management server 30.

In some embodiments, a recommendation response to a query reported from the management server 30 may be received. In this case, the target model may be updated based on the conversation pair including the reported query and the recommended response. The method of updating the target model may be a method of adding rules or scenarios generated based on the conversation pair, a method of performing additional learning on the conversation pair, etc., but the technical scope of the present disclosure is limited thereto. It does not work.

In step S450, in response to the determination that the feedback is positive, a target model may be updated based on the conversation pair including the received query and the generated response.

So far, with reference to FIG. 15, a method of updating a conversation model has been described in some embodiments of the present disclosure. According to the above-described method, the target model may be automatically updated gradually based on user feedback. Accordingly, the human cost of managing the conversation model can be reduced, and the convenience of the manager can be further improved.

16 and 17 are diagrams for explaining a conversation model update method according to some other embodiments of the present disclosure. In particular, FIGS. 16 and 17 illustrate a method of updating the conversation model in the management server 30.

As illustrated in FIG. 16, the method for updating the conversation model starts at step S500 of receiving historical data from one or more chatbot servers 20. As described above, the history data may include query, response, and feedback information. For example, the history data may include unanswered queries and query and response data where feedback is predicted negatively. Hereinafter, in order to provide convenience for understanding, it is assumed that the history data is received from the first chatbot server 20-1 to continue the description.

In step S510, it is determined whether there is an unanswered query or a negative feedback query in the historical data.

In step S520, in response to the presence determination, a recommendation response to the queries is input. For example, the recommendation response may be input by an administrator. In some other embodiments, recommendation responses may be generated automatically by analyzing accumulated historical data. For example, a response in which feedback is very positive among responses to similar queries may be generated as a recommended response.

In step S530, at least a part of the plurality of service domains is recommended as an update target. In detail, based on the similarity between the first service domain associated with the first chatbot server 20-1 and other service domains, the at least a portion may be recommended as an update target.

In this step S530, a specific method of determining an update target may vary according to an embodiment.

In some embodiments, the update target includes the first service domain, and may further include a service domain having a similarity with the first service domain that is greater than or equal to a reference value.

In some embodiments, if the number of service domains having a similarity or higher than the threshold value is greater than or equal to a threshold value among service domains located in the same layer (exactly, among the service domains having the same upper domain), the upper service domain of the service domains is subject to update Can be recommended as To provide convenience of understanding, the present embodiment will be described in detail with reference to FIG. 17. FIG. 17 illustrates a case where the first service domain associated with the first chatbot server 20-1 is the B hotel domain 151, the reference value is "80%", and the threshold value is "1/2".

Referring to the hierarchical structure 150 illustrated in FIG. 17, among service domains 151 to 155 of the same layer as the B hotel domain 151, two service domains 151 and 153 having a similarity or higher than the reference value are counted. Since the counted number is more than 1/2 of the total (three), the upper service domain 159 may be recommended as an update target. The reason for recommending the upper service domain 159 as an update target is that the higher the number of surrounding service domains (eg 153, 155) with high similarity, the higher the likelihood that the historical data obtained from the B hotel domain 151 corresponds to the common part. Because.

In the example shown in FIG. 17, if the upper service domain 159 is finally determined to be updated, the conversation model of the upper service domain 159 is updated, and the conversation model of the lower service domains 151 to 155 is also updated. It can be updated in bulk depending on the conversation model. This is because the upper dialogue model and the lower dialogue model are inherited.

In some other embodiments, an update target may be recommended based further on the performance of the conversation model associated with each service domain. For example, conversation models of a service domain having similarity with the first service domain having a reference value or more are determined as candidate conversation models, and among the candidate models, a model having a performance below a threshold may be recommended as an update target. As described above, the performance of the conversation model may be determined based on the number of scenarios associated with each conversation model, the number of unanswered queries, and user feedback on the answered query. However, the technical scope of the present disclosure is not limited thereto.

In steps S540 and S550, when correction for an update target is required, the target correction may be performed. For example, a correction may be made such that only some service domains among service domains recommended as update targets are selected as actual update targets, or service domains not recommended are added. Steps S540 and S550 may be performed by the administrator to improve the accuracy of selection of the update target. Of course, if necessary, steps S540 and S550 may be omitted.

In step S560, the conversation model associated with the service domain determined to be updated is updated.

So far, a method of updating a conversation model according to various embodiments of the present disclosure has been described with reference to FIGS. 16 and 17. According to the above-described method, automatic recommendation for an update target (range) may be made based on indicators such as similarity and performance between service domains. Therefore, the human cost for managing the conversation model is reduced, and the convenience of the manager can be further improved.

Hereinafter, an exemplary computing device 200 capable of implementing a device (e.g. chatbot server 20, management server 30) according to various embodiments of the present disclosure will be described with reference to FIG. 18.

As shown in FIG. 18, the computing device 200 includes a memory (for loading) a computer program performed by one or more processors 210, a bus 250, a communication interface 270, and a processor 210. 230, and a storage 290 for storing the computer program 291. However, only components related to the exemplary embodiment of the present disclosure are illustrated in FIG. 18. Accordingly, a person skilled in the art to which the present disclosure belongs may recognize that other general-purpose components may be further included in addition to the components illustrated in FIG. 18.

The processor 210 controls the overall operation of each component of the computing device 200. The processor 210 includes a CPU (Central Processing Unit), a Micro Processor Unit (MPU), a Micro Controller Unit (MCU), a Graphic Processing Unit (GPU), or any type of processor well known in the art of the present disclosure. Can be. Also, the processor 210 may perform operations on at least one application or program for executing the method according to embodiments of the present disclosure. Computing device 200 may include one or more processors.

The memory 230 stores various data, commands and/or information. The memory 230 may load one or more programs 291 from the storage 290 to execute a method/operation according to various embodiments of the present disclosure. The memory 230 may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

The bus 250 provides communication functions between components of the computing device 200. The bus 250 may be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

The communication interface 270 supports wired and wireless Internet communication of the computing device 200. Also, the communication interface 270 may support various communication methods other than Internet communication. To this end, the communication interface 270 may include a communication module well known in the art of the present disclosure.

The storage 290 may store the one or more programs 291 and various data non-temporarily. The various data may include a conversation model, various history data, and the like. The storage 290 is well-known in the art of non-volatile memory such as read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, or the present disclosure. And any known form of computer-readable recording media.

Computer program 291 may include one or more instructions that, when loaded into memory 230, cause processor 210 to perform methods/operations in accordance with various embodiments of the present disclosure. That is, the processor 210 may perform the methods/operations according to various embodiments of the present disclosure by executing the one or more instructions.

For example, the computer program 291 determines a target service domain to which a target chatbot is applied among a plurality of hierarchically defined service domains, and obtains a first conversation model for a higher service domain of the target service domain, To inherit the first conversation model to generate a second conversation model of the target service domain, and to perform an operation of providing a chatbot service for the target service domain through the target chatbot using the second conversation model It may include instructions. In this case, the chatbot server 20 according to some embodiments of the present disclosure may be implemented through the computing device 200.

For another example, the computer program 291 may receive a query from a user terminal through a target chatbot, and determine whether to respond to the query based on a conversation model of a target service domain to which the target chatbot is applied, In response to the determination that the response is not possible, an operation of selecting a reference conversation model from among the plurality of conversation models based on the similarity between service domains and a response generated using the reference conversation model through the target chatbot It may include instructions to perform the provided operation. In this case, the chatbot server 20 according to some other embodiments of the present disclosure may be implemented through the computing device 200.

For another example, the computer program 291 obtains query data including an unanswered query from a first chatbot utilized in a first service domain among a plurality of service domains, and recommends a response to the query data. Using an input operation, an operation of recommending at least a part of the plurality of service domains as an update target based on the similarity between the first service domain and another service domain, and using update data including the query data and the recommendation response And instructions for performing an operation of updating a conversation model associated with the recommended service domain. In this case, the management server 20 according to some embodiments of the present disclosure may be implemented through the computing device 200.

So far, an exemplary computing device 200 capable of implementing an apparatus according to various embodiments of the present disclosure has been described with reference to FIG. 18.

So far, various embodiments of the present disclosure and effects according to the embodiments have been described with reference to FIGS. 1 to 18. Effects according to the technical spirit of the present disclosure are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

The technical idea of the present disclosure described so far with reference to FIGS. 1 to 18 may be embodied as computer readable codes on a computer readable medium. The computer-readable recording medium may be, for example, a removable recording medium (CD, DVD, Blu-ray Disc, USB storage device, removable hard disk), or a fixed recording medium (ROM, RAM, computer-equipped hard disk). Can. The computer program recorded on the computer-readable recording medium may be transmitted to another computing device through a network such as the Internet and installed on the other computing device, and thus used on the other computing device.

In the above, even if all components constituting the embodiments of the present disclosure are described as being combined or operated as one, the technical spirit of the present disclosure is not necessarily limited to these embodiments. That is, within the scope of the present disclosure, all of the components may be selectively combined and operated.

Although the operations are shown in a specific order in the drawings, it should not be understood that the operations must be performed in a specific order or in a sequential order, or all illustrated operations must be executed to obtain a desired result. In certain situations, multitasking and parallel processing may be advantageous. Moreover, the separation of various configurations in the above-described embodiments should not be understood as such a separation is not necessarily necessary, and the described program components and systems may generally be integrated together into a single software product or packaged into multiple software products. It should be understood that there is.

Although the embodiments of the present disclosure have been described with reference to the accompanying drawings, those of ordinary skill in the art to which the present disclosure pertains may implement the present disclosure in other specific forms without changing its technical spirit or essential features. You can understand that there is. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. The scope of protection of the present disclosure should be interpreted by the claims below, and all technical spirits within the scope equivalent thereto should be interpreted as being included in the scope of the technical spirits defined by the present disclosure.

Claims (22)

A method for providing a chatbot service performed by a computing device,
Determining a target service domain to which a target chatbot is applied among a plurality of hierarchically defined service domains;
Obtaining a first conversation model for an upper service domain of the target service domain;
Generating a second conversation model of the target service domain by inheriting the first conversation model; And
And providing a chatbot service for the target service domain through the target chatbot using the second conversation model.
How to provide chatbot service. According to claim 1,
The step of providing the chatbot service,
Receiving a query from a user terminal;
Generating a response to the query using the second conversation model; And
Characterized in that it comprises the step of providing the generated response to the user terminal,
How to provide chatbot service. According to claim 2,
Generating a response to the query,
Extracting an intent included in the query;
Determining a scenario meeting the intention through the second conversation model; And
Characterized in that it comprises the step of generating the response based on the determined scenario,
How to provide chatbot service. According to claim 1,
The step of providing the chatbot service,
Receiving a query from a user terminal;
Determining whether a response to the query is possible based on the second conversation model;
In response to determining that the response is not possible, selecting a reference service domain from the plurality of service domains based on the similarity between service domains; And
Characterized in that it comprises the step of providing a response generated using the third conversation model associated with the reference service domain to the user terminal,
How to provide chatbot service. According to claim 4,
The step of selecting the reference service domain,
And selecting a service domain that satisfies the criterion similarity allowed for the received query as the reference service domain,
How to provide chatbot service. According to claim 4,
The step of selecting the reference service domain,
Selecting a plurality of candidate service domains from among a plurality of service domains based on the similarity;
Generating a plurality of responses using a conversation model of each of the plurality of candidate service domains;
Evaluating the quality of each of the plurality of responses; And
And selecting the reference service domain based on the evaluation result.
How to provide chatbot service. According to claim 4,
Obtaining feedback on the generated response from the user terminal; And
And adjusting the similarity to the reference service domain based on the feedback.
How to provide chatbot service. According to claim 4,
The similarity is,
Characterized in that it is calculated based on the similarity between the specification information of the target service domain and the specification information of another service domain,
How to provide chatbot service. A method for providing a chatbot service using a plurality of conversation models utilized in a plurality of service domains on a computing device,
Receiving a query from a user terminal through a target chatbot;
Determining whether a response to the query is possible based on a conversation model of a target service domain to which the target chatbot is applied;
In response to determining that the response is not possible, selecting a reference conversation model from among the plurality of conversation models based on the similarity between service domains; And
And providing a response generated using the reference conversation model through the target chatbot.
How to provide chatbot service. The method of claim 9,
The step of selecting the conversation model for reference,
And selecting a conversation model of a service domain that satisfies a criterion similarity allowed for the received query among the plurality of service domains as the reference conversation model.
How to provide chatbot service. The method of claim 9,
The step of selecting the conversation model for reference,
Selecting a plurality of candidate service domains from among a plurality of service domains based on the similarity;
Generating a plurality of responses using a conversation model of each of the plurality of candidate service domains;
Evaluating the quality of each of the plurality of responses; And
And selecting the dialogue model for reference based on the evaluation result.
How to provide chatbot service. The method of claim 9,
Obtaining feedback on the generated response from the user terminal; And
And adjusting the similarity of service domains associated with the reference conversation model based on the feedback.
How to provide chatbot service. The method of claim 12,
The step of obtaining the feedback,
And if the feedback is not received within a specified time period, predicting feedback for the generated response.
How to provide chatbot service. The method of claim 12,
In response to determining that the feedback is positive, further comprising updating a conversation model of the target service domain based on the conversation pair comprising the received query and the generated response,
How to provide chatbot service. The method of claim 12,
In response to determining that the feedback is negative, reporting the received query to a management server;
Receiving a recommendation response to the received query from the management server; And
And updating a conversation model of the target service domain based on the conversation pair including the received query and the recommended response.
How to provide chatbot service. The method of claim 9,
The similarity is,
Characterized in that it is calculated based on the similarity between the first specification information of the target service domain and the second specification information of the other service domain,
How to provide chatbot service. The method of claim 16,
The first specification information and the second specification information are specification information associated with a specific intention,
The similarity is calculated for each intention,
The step of selecting the conversation model for reference,
Extracting the intention included in the query; And
And selecting the dialogue model for reference based on the similarity to the extracted intention,
How to provide chatbot service. A method of providing a chatbot service using a plurality of chatbots utilized in a plurality of service domains on a computing device,
Obtaining query data including an unanswered query from a first chatbot utilized in a first service domain among the plurality of service domains;
Receiving a recommendation response to the query data;
Recommending at least a portion of the plurality of service domains as an update target based on the similarity between the first service domain and other service domains; And
And updating a conversation model associated with the recommended service domain by using the update data including the query data and the recommendation response.
How to provide chatbot service. The method of claim 18,
The query data is characterized in that it further comprises a query received negative user feedback,
How to provide chatbot service. The method of claim 18,
The recommended step for the update target,
Recommending at least some of the above based on the performance of the conversation model associated with each service domain,
Characterized in that the performance of the conversation model is determined based on the number of scenarios associated with each conversation model,
How to provide chatbot service. The method of claim 18,
The recommended step for the update target,
Recommending at least some of the above based on the performance of the conversation model associated with each service domain,
The performance of the conversation model is determined based on the number of unanswered queries and user feedback on the answered queries,
How to provide chatbot service. The method of claim 18,
The plurality of service domains are defined in a hierarchical structure,
In the step of recommending the update target,
Counting the number of service domains in which the similarity between the service domains is greater than or equal to a reference value in a first layer in which the first service domain is located; And
And in response to a determination that the counted number is greater than or equal to a threshold, recommending a second service domain located in an upper layer of the first layer as the update target,
The updating step,
Updating a first conversation model associated with the second service domain using the update data; And
And updating the conversation model associated with a sub-service domain of the second service domain based on the updated first conversation model.
How to provide chatbot service.

Images