Conversational artificial intelligence system and method for providing integrated question-and-answer service

Abstract

The present invention discloses a conversational artificial intelligence system. More specifically, the present invention relates to a conversational artificial intelligence system for providing an integrated question-and-answer service, which can process questions and answers on various types of subjects either at a casual conversation level or based on specialized knowledge. The present invention, according to an embodiment, enables seamless integration of casual conversations and legal knowledge-based question-and-answer by providing a conversational artificial intelligence interface that delivers both everyday conversations and explainable legal-domain question-and-answer services in an integrated manner, and thus can offer more natural responses and enhanced immersion in the conversation for users of the system .

Description

Conversational artificial intelligence system and method for providing complex question-answering services

[0001] The present invention relates to a conversational artificial intelligence system, and more particularly, to a conversational artificial intelligence system and method that provide a complex question-answering service capable of processing question-answering on various types of topics at a daily level or based on specialized knowledge.

[0002] Recently, with the emergence of large-scale language models based on artificial intelligence technologies such as ChatGPT, various forms of everyday conversation services based on these models, as well as specialized conversation services that provide question-and-answer services utilizing specialized domain knowledge, are being commercialized.

[0003] However, currently commercialized conversation services are limited to providing only ordinary, everyday conversations such as personal stories, or to providing only knowledge-based conversations specialized in specialized domains such as technology.

[0004] As large-scale language models achieve high performance in natural language processing at current technological levels, legal technology, which utilizes language models in fields requiring specialized knowledge such as law, is rapidly developing. However, when large-scale language models are used for legal question-and-answer, the so-called hallucination problem, where inaccurate information is conveyed fluently, is occurring, raising concerns about reliability and accuracy.

[0005] The aforementioned hallucination problem refers to a situation where the output or prediction of an artificial intelligence system deviates significantly from reality or expected normal results.

[0006] Accordingly, the present invention seeks to develop a system that provides a combination of everyday conversation and legal domain question and answering, and to implement an explainable and highly reliable explainable artificial intelligence (XAI) system by providing a basis for answers, especially in the case of knowledge-based answers.

[0007] (Patent Document 1) Patent Registration No. 10-2574645 (Publication Date: September 7, 2023)

[0008] The present invention has the task of implementing a conversational artificial intelligence system capable of processing complex conversations by integrating a single system for everyday conversation or knowledge-based question-and-answer, which was previously implemented in a single form.

[0009] In particular, the present invention aims to implement a system that can integrate two tracks of systems into one by classifying user inputs requiring knowledge-based explanations in order to integrate daily conversation and a knowledge-based question-answering system. In the case of the daily conversation system, a process of identifying the user's intention and generating an appropriate answer by utilizing a large-scale language model is provided as an end-to-end technology, and the legal knowledge-based question-answering system is characterized by improving the accuracy and reliability of artificial intelligence technology by implementing a system that searches for and provides a basis for an answer in order to solve the hallucination problem of the large-scale language model.

[0010] In order to solve the above-described problem, a conversational artificial intelligence system providing a complex question-answering service according to one embodiment of the present invention is a system providing a question-answering service using a first language model provided by an external system, the system including an interface unit for receiving a question for a conversational service from a user terminal connected through an information and communication network and returning a response, a first prompt having a plurality of domain lists set, a search unit for inputting a question entered through the interface unit according to the first prompt into the first language model and receiving a result of a judgment as to whether a knowledge-based explanation is necessary and classifying the question as a daily conversation or a knowledge-based conversation, and an information database for storing a plurality of related pieces of information regarding a knowledge-based conversation, wherein the search unit can proceed with a conversation process in an end-to-end manner with the first language model for a question classified as a daily conversation, and can proceed with a conversation process in one or more pipeline manners with reference to the information database for a question classified as a knowledge-based conversation.

[0011] The above search engine, in the case of a query classified as a daily conversation, can input a detailed query and output a response including a second prompt including multiple topics related to daily life to the first language model and provide the response to the user terminal in multiple steps.

[0012] The above daily conversation may include questions and answers about the current weather and the user's schedule.

[0013] The above search engine can update the second prompt to reflect the results of the request in the response when the user requests to add or change the schedule.

[0014] The above search device, in the case of a query classified as a specialized domain conversation, searches the information database using a second language model to extract one or more related pieces of information matching the query, and inputs a detailed query to the first language model through a third prompt including the extracted related information, outputs a response, and provides the result to a user terminal.

[0015] The above second language model can be fine-tuned using the detailed query and related information corresponding to the detailed query as a dataset.

[0016] The above knowledge-based conversation may include legal domain related questions and answers.

[0017] The above search engine can update the third prompt to proceed with the conversation process through a first pipeline that processes questions and answers regarding the articles of incorporation in a machine-readable manner and a second pipeline that processes them in a legal professional advice-generating manner.

[0018] In addition, in order to solve the above-mentioned problem, a method for providing a composite question-answering service according to an embodiment of the present invention is a method for providing a question-answering service using a first language model provided by an external system by a conversational artificial intelligence system, the method including: (a) a step of receiving a question for a conversational service from a user terminal connected through an information and communication network, (b) a step of inputting a question input through a first prompt in which a plurality of domain lists are set into the first language model, (c) a step of receiving a judgment result on whether a knowledge-based explanation is necessary from the first language model and classifying the question as a daily conversation or a knowledge-based conversation, and (d) a step of proceeding with a conversation process in an end-to-end manner with the first language model for a query classified as a daily conversation, or proceeding with a conversation process in one or more pipeline manners by referencing an information database for a query classified as a knowledge-based conversation.

[0019] The step (d) above may include, in the case of a query classified as a daily conversation, (d1) a step of inputting a detailed query into the first language model through a second prompt including multiple topics related to daily life, (d2) a step of outputting a response to the detailed query and providing it to the user terminal, and (d3) a step of repeatedly performing steps (d1) and (d2) until the conversation ends.

[0020] The above daily conversation may include questions and answers about the current weather and the user's schedule.

[0021] The above step (d1) may include a step of updating the second prompt to reflect the result of the request in the response when the user requests to add or change the schedule.

[0022] The step (d) above may include, in the case of a query classified as a specialized domain conversation, (d4) a step of searching the information database using a second language model to extract one or more related pieces of information matching the query, (d5) a step of inputting a detailed query into the first language model through a third prompt including the extracted related information, (d6) a step of outputting a response from the first language model and providing the response to a user terminal, and (d7) a step of repeating steps (d4) to (d6) until the conversation ends.

[0023] Before the above step (d), a step of fine-tuning the second language model using the detailed query and related information corresponding to the detailed query as a dataset may be included.

[0024] The above knowledge-based conversation may include legal domain related questions and answers.

[0025] The above step (d4) may include a step of updating the third prompt to proceed with the conversation process by dividing the first pipeline into a machine-readable process for questions and answers regarding the articles of incorporation and a second pipeline into a legal professional advice generation process.

[0026] According to an embodiment of the present invention, by providing an artificial intelligence conversation interface that provides a combination of everyday conversation and explainable legal domain question and answer, it is possible to provide a combination of everyday conversation and legal knowledge-based question and answer in a smooth and smooth manner, thereby increasing the conversational immersion and naturalness of responses of users using the system.

[0027] In addition, the present invention has the effect of providing an end-to-end technology with high accuracy while generating an answer to a user's input end-to-end by applying a large-scale language model prompting technology in a part that provides everyday conversation.

[0028] In addition, the present invention has the effect of providing explainability by presenting accurate and reliable answers with valid grounds in the legal domain by combining the technology for searching for judgment grounds in the legal knowledge-based answer generation section with the technology for prompting a large-scale language model.

[0029] FIG. 1 is a diagram schematically illustrating the entire connection structure of an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0030] FIG. 2 is a drawing showing in detail the structure of an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0031] FIG. 3 is a diagram illustrating a method for providing a composite question-answering service according to an embodiment of the present invention.

[0032] FIG. 4 is a diagram illustrating a classification prompt for classifying whether a knowledge-based explanation is needed in an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0033] FIG. 5 and FIG. 6 are diagrams illustrating conversation content using an interactive artificial intelligence system that provides a complex question-and-answer service according to an embodiment of the present invention.

[0034] FIG. 7 is a diagram illustrating a learning data set and the number of data of a searcher used in an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0035] FIG. 8 is a diagram illustrating a prompt for generating a knowledge-based dialogue for question-answering on an automatic review problem of articles of incorporation used in an interactive artificial intelligence system providing a complex question-answering service according to an embodiment of the present invention.

[0036] The present invention as described above will be described in detail through the attached drawings and examples.

[0037] It should be noted that the technical terms used in the present invention are used merely to describe specific embodiments and are not intended to limit the present invention. Furthermore, unless specifically defined otherwise herein, the technical terms used herein should be interpreted as having a meaning generally understood by those skilled in the art to which the present invention pertains, and should not be interpreted in an excessively broad or narrow sense. Furthermore, if a technical term used herein is incorrect and fails to accurately express the spirit of the present invention, it should be replaced with a technical term that can be correctly understood by those skilled in the art. Furthermore, general terms used herein should be interpreted according to their dictionary definitions or according to the context, and should not be interpreted in an excessively narrow sense.

[0038] Additionally, singular expressions used in the present invention include plural expressions unless the context clearly dictates otherwise. In the present invention, terms such as "consist of" or "include" should not be construed to necessarily include all of the components or steps described in the invention, and should be construed to mean that some of the components or steps may not be included, or that additional components or steps may be included.

[0039] Additionally, terms including ordinal numbers, such as "first" and "second," used in the present invention may be used to describe components, but the components should not be limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present invention, a first component could be referred to as a "second component," and similarly, a second component could also be referred to as a "first component."

[0040] Furthermore, the various techniques described herein may be implemented with hardware or software, or, where appropriate, with a combination of both. Terms such as "unit" and "system" as used herein may likewise be treated as equivalent to computer-related entities, i.e., hardware, a combination of hardware and software, software, or software in execution. Furthermore, each function implemented in the system of the present invention may be configured as a modular program, and may be recorded in a single physical memory, or may be distributed and recorded between two or more memories and storage media.

[0041] Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the attached drawings. Regardless of the drawing numbers, identical or similar components are given the same reference numbers and redundant descriptions thereof will be omitted.

[0042] In the following description, the term "conversational artificial intelligence system providing a complex question-answering service" according to an embodiment of the present invention may be described interchangeably as "conversational artificial intelligence system" or "system."

[0043] Hereinafter, an interactive artificial intelligence system providing a composite question-answering service according to an embodiment of the present invention will be described with reference to the drawings.

[0044] FIG. 1 is a diagram schematically illustrating the entire connection structure of an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0045] Referring to FIG. 1, a conversational artificial intelligence system (100) providing a complex question-answering service according to an embodiment of the present invention is a system that provides a question-answering service using a large-scale language model provided by an external system (200), and can provide an artificial intelligence-based conversational service in response to a service request from a user terminal (10) connected through an information and communications network.

[0046] In particular, the conversational artificial intelligence system (100) according to an embodiment of the present invention has a feature that, when a user input of a user terminal (10) is received, if it is determined that a knowledge-based explanation is necessary, it proceeds with a knowledge-based question and answer process, otherwise it proceeds with a different process according to each type of conversation, thereby enabling a conversation appropriate to the user's intention.

[0047] The user terminal (10) is a terminal possessed and used by a user who wishes to conduct a conversation with an AI chatbot based on artificial intelligence online through the system (100) of the present invention, and may be a smartphone or tablet PC, etc., and can input the content of a query and transmit it to the system (100), and receive and display a response therefrom.

[0048] The conversation service with such a system (100) can be performed based on a web browser or a dedicated application.

[0049] The conversational artificial intelligence system (100) can provide a complex question-and-answer service at the request of one or more user terminals (10), and has a feature of analyzing the content of a question input from a user terminal (10) to enable a conversation appropriate to the user's intention.

[0050] The conversational AI system (100) according to an embodiment of the present invention can determine whether the topic of the user's requested conversation is a general, everyday conversation about daily life or a conversation based on specialized knowledge, and perform the conversation process in a separate route. Here, everyday conversation refers to a conversation that includes questions and answers about the current weather and the user's schedule, while knowledge-based conversation refers to a conversation that includes questions and answers related to the legal domain.

[0051] In particular, the conversational AI system (100) according to an embodiment of the present invention can improve the accuracy of its answers by dividing legal domain questions into subproblems and solving them, as these require complex logical reasoning as knowledge-based conversations. For each subproblem, information related to the user's question is retrieved from the onboard information database, and then appropriate prompting to an external, large-scale language model is applied to generate highly accurate and reliable results.

[0052] In addition, the interactive artificial intelligence system (100) of the present invention is characterized by being equipped with a searcher of an original structure for accurate information retrieval, through which a Korean dictionary learning language model can be fine-tuned using data matching information about a question of a detailed problem.

[0053] Additionally, according to an embodiment of the present invention, through the process of analyzing a query and deriving a response, it is possible to provide explainability by presenting the retrieved information along with an accurate answer to a legal issue with relevant evidence.

[0054] A detailed description of an interactive artificial intelligence system providing a composite question-answering service according to an embodiment of the present invention for implementing the aforementioned function is provided below.

[0055] In addition, the external system (200) is a system that provides a large-scale conversational artificial intelligence model, for example, ChatGPT, for providing the complex question-answering service of the present invention, and can provide answers to questions of the conversational artificial intelligence system (100) through an API.

[0056] Through this, the conversational artificial intelligence system (100) of the present invention generates responses by utilizing a large-scale conversational artificial intelligence model provided in conjunction with an external system (200), but enables conversations to be performed through different pipelines for everyday conversations and knowledge-based conversations.

[0057] According to the structure described above, the conversational artificial intelligence system providing a complex question-answering service according to an embodiment of the present invention implements a conversational service based on large-scale conversational artificial intelligence, analyzes a query provided by a user to determine whether it is a knowledge-based conversation, and thereby derives and provides a more appropriate response to the query.

[0058] Hereinafter, an interactive artificial intelligence system providing a composite question-answering service according to an embodiment of the present invention will be described in detail with reference to the drawings.

[0059] FIG. 2 is a drawing showing in detail the structure of an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0060] Referring to FIG. 2, a conversational artificial intelligence system (100) providing a complex question-answering service according to an embodiment of the present invention includes an interface unit (110) for receiving a question for a conversational service from a user terminal (10) connected through an information and communication network and returning a response, a classification prompt (121) in which a plurality of domain lists are set, a searcher (120) for inputting a question entered through the interface unit (110) into a large-scale language model (230) according to the classification prompt (121) and receiving a judgment result on whether a knowledge-based explanation is necessary to classify the question as an everyday conversation or a knowledge-based conversation, and an information database (140) for storing a plurality of related pieces of information regarding a knowledge-based conversation, and the searcher (120) can proceed with a conversation process in an end-to-end manner with the large-scale language model (230) for a question classified as an everyday conversation, and can proceed with a conversation process in one or more pipeline manners with reference to the information database (140) for a question classified as a knowledge-based conversation.

[0061] The interface unit (110) performs data communication with a user terminal (10) connected to the system (100), and in particular, can receive a query for conversation progress and send a response thereto to the user terminal (10).

[0062] When a user's query is input through the interface unit (110), the search engine (120) inputs the query into a large-scale language model (230) provided by an external system (200) through a classification prompt (121), which is the first prompt for classification, thereby receiving a judgment result as to whether the subject of the input query is an everyday conversation or a knowledge-based conversation.

[0063] And, in the case of a query classified as a daily conversation, the search engine (120) includes a daily conversation prompt (123), which is a second prompt for daily conversation that includes multiple topics related to daily life in a large-scale language model (230), and through this, a detailed query can be input and a response can be output and provided to the user terminal (10) in multiple steps.

[0064] In addition, when a user requests to add or change a schedule, the search engine (120) updates the daily conversation prompt (123) to reflect the result of the request in the response, thereby reflecting the necessary information and providing a response thereto to the user terminal (10).

[0065] In addition, the search engine (120) can search the information database (140) using the pre-learning language model (130), which is a second language model, to extract one or more related pieces of information matching the query when the current conversation is classified as a knowledge-based conversation requiring specialized knowledge, for example, a legal domain conversation. In addition, the search engine (120) can include a knowledge-based conversation prompt (125), which is a third prompt for a knowledge-based conversation including the extracted related information, and input a detailed query into the large-scale language model (230) through this, thereby outputting a response thereto and providing it to the user terminal (10).

[0066] Here, the pre-learning language model (130) described above may be a fine-tuned learning model learned using a dataset including detailed queries and related information corresponding to the detailed queries.

[0067] In particular, in the case of a legal domain-based conversation, the search engine (120) can update the knowledge-based conversation prompt (125) to proceed with the conversation process at least through the first pipeline that processes questions and answers regarding the articles of incorporation in a machine-reading manner and the second pipeline that processes them in a legal professional advice generation manner.

[0068] The pre-trained language model (130) is a second language model installed in the system (100) of the present invention and is a learning model for conducting knowledge-based conversations. The system (100) can extract relevant information for queries classified as knowledge-based conversations by the knowledge-based conversation prompt (125). For example, if a conversation requires specialized knowledge, such as a legal domain, the pre-trained language model (130) can retrieve questions related to the articles of incorporation. For this purpose, the pre-trained language model can be pre-trained using a dataset for automatic review of articles of incorporation.

[0069] The information database (140) can store relevant information regarding various specialized knowledge for implementing knowledge-based conversations. In particular, the information database (140) can be referenced by the pre-trained language model (130) to provide relevant information necessary for detailed queries to the large-scale language model (230), and can be input into the large-scale language model (230) via the knowledge-based conversation prompt (125).

[0070] The relevant information provided in this information database (140) may be information that has been reviewed in advance by a person with relevant expertise, such as a lawyer.

[0071] In particular, according to an embodiment of the present invention, the system (100) retrieves information related to a user's question from an information database (140) regarding detailed issues included in the query, and then generates a result with high accuracy and reliability through appropriate prompting to a large-scale language model (230).

[0072] Hereinafter, a method for providing a composite question-answering service according to an embodiment of the present invention will be described in detail with reference to the drawings.

[0073] Figure 3 is a diagram illustrating a method for providing a composite question-and-answer service according to an embodiment of the present invention. In the following description, the executor of each step is the interactive artificial intelligence system and components of the present invention, even if not otherwise specified.

[0074] Referring to FIG. 3, a method for providing a composite question-answering service according to an embodiment of the present invention first allows a user who wishes to use the question-answering service of the present invention to access an interactive artificial intelligence system using his or her user terminal and input a question for conversation (S100). Then, the system may input the input question through a classification prompt into a large-scale language model provided by an external system, for example, ChatGPT, and request the system to classify the question according to topic (S110).

[0075] Accordingly, the large-scale language model determines whether the conversation content included in the query belongs to an everyday conversation or a knowledge-based conversation through a classification prompt and returns the classification result to the system (S120).

[0076] Based on the classification results (S125), if the input query is determined to be a daily conversation, the system inputs a detailed query for subsequent conversation progression into a large-scale language model through a daily conversation prompt (S130), receives a response thereto, and provides it to the user terminal (S140), thereby allowing the conversation to proceed.

[0077] Here, the daily conversation may include questions and answers about the current weather and the user's schedule, and as the conversation progresses, when the user requests additions and changes to their schedule, the daily conversation prompt may be updated to reflect the results of the requests in the response.

[0078] This can be viewed as a way to conduct a conversation through a single pipeline between a large-scale language model and the system, and the system can repeat steps S130 to S140 described above until the conversation ends.

[0079] In addition, if the input query is determined to be a knowledge-based conversation requiring specialized knowledge based on the classification result of the aforementioned query content (S125), the system can extract relevant information related to the conversation topic using the pre-learning (second) language model installed (S150), input a detailed query for conversation progress into a large-scale language model using the relevant information extracted through the third prompt (S160), and receive a response thereto and provide it to the user terminal (S170), thereby allowing the conversation to proceed.

[0080] Here, the knowledge-based conversation may include legal domain-related questions and responses, and prior to step S150, may include a step of fine-tuning a pre-learning language model using detailed queries and related information corresponding to the detailed queries as a dataset.

[0081] This is a conversational process where the system, including a large-scale language model, divides multiple detailed queries into pre-trained language models and receives queries and responses. This process can be viewed as a conversational process conducted through multiple pipelines. Furthermore, the system can repeat steps S150 through S170 until the conversation ends.

[0082] In particular, the aforementioned knowledge-based conversation may include a step of updating a knowledge-based conversation prompt to proceed with the conversation process by dividing the legal domain conversation into a first pipeline that processes questions and answers about articles of incorporation in a machine-readable manner and a second pipeline that processes them in a legal professional advice-generating manner.

[0083] Hereinafter, the technical idea of ​​the present invention will be described in detail through examples of prompts and dialogue content of an interactive artificial intelligence system that provides a complex question-and-answer service according to an embodiment of the present invention.

[0084] FIG. 4 is a diagram illustrating a classification prompt for classifying whether a knowledge-based explanation is needed in an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0085] FIG. 4 illustrates the structure of a classification prompt for classifying whether a knowledge-based explanation is needed, used in an embodiment of the present invention, and shows a structure set up so that a large-scale language model can select one of the domains of a conversation that is a candidate for the classification prompt.

[0086] According to this structure, a large-scale language model can classify various topics in everyday conversation, such as weather, schedule, and contact information, and derive and respond appropriate answers when the user's input is received from the system and classified as input corresponding to everyday conversation.

[0087] At this time, questions and answers are processed in detailed steps based on the topic, and can be resolved in an end-to-end manner using a single language model, rather than the conventional pipeline method that required the use of additional models at each stage according to the progress of the conversation.

[0088] On the other hand, if a large-scale language model classifies the user's input as input requiring a knowledge-based explanation, the necessary information for the question can be retrieved and an answer can be provided according to the information retrieval-based question answering pipeline method.

[0089] FIG. 5 and FIG. 6 are diagrams illustrating conversation content using an interactive artificial intelligence system that provides a complex question-and-answer service according to an embodiment of the present invention.

[0090] Referring to Figure 5, the system according to an embodiment of the present invention provides a question-and-answer service, categorized as a daily conversation, illustrating the conversation. The system provides accurate answers to users' inquiries about today's weather and can appropriately execute user instructions for adding or changing schedules.

[0091] Additionally, Figure 6 illustrates a conversation categorized as a legal domain knowledge-based conversation through a question-and-answer service provided by a system according to an embodiment of the present invention. The system can derive and provide appropriate responses to questions regarding automatic review of articles of incorporation through prompting from a pre-learning model.

[0092] In particular, the problem of automatic review of articles of incorporation can be handled by a complex inference step of machine reading the articles of incorporation to check the attorney's checklist, and generating attorney advice regarding the question-answer pairs of the checklist to see if the articles of incorporation are written legally correctly. To solve this, according to an embodiment of the present invention, the problem of machine reading the articles of incorporation can be divided into two sub-queries of solving the problem of machine reading and generating attorney advice, and two information retrieval-based question-answer pipelines can be used to conduct a conversation.

[0093] FIG. 7 is a diagram illustrating a learning data set and the number of data of a searcher used in an interactive artificial intelligence system that provides a complex question-answering service according to an embodiment of the present invention.

[0094] Referring to Figure 7, the system can use the articles and questions from the Articles of Incorporation Automatic Review dataset to train a search engine that retrieves questions related to the articles of incorporation, given the articles of incorporation. For example, to retrieve relevant commercial law provisions based on the articles of incorporation review results generated from detailed questions, a pre-trained language model can be trained using the questions and commercial law items from the Articles of Incorporation Automatic Review dataset and responses can be derived.

[0095] FIG. 8 is a diagram illustrating a prompt for generating a knowledge-based dialogue for question-answering on an automatic review problem of articles of incorporation used in an interactive artificial intelligence system providing a complex question-answering service according to an embodiment of the present invention.

[0096] Referring to Figure 8, a knowledge-based dialogue prompt for question-and-answering on an automated Articles of Incorporation review problem used in the system of the present invention is illustrated. The knowledge-based dialogue prompt for solving the Articles of Incorporation review machine reading problem corresponding to Detailed Question 1 may include a question retrieved from a search engine, the corresponding Articles of Incorporation document, and instructions for generating an answer to the question.

[0097] Additionally, the knowledge-based dialogue prompt for generating legal advice for subquestion 2 can generate legal advice by incorporating the question, the answer generated from subquestion 1, and the commercial law provisions retrieved from the search engine. By instructing the user to generate advice based on the commercial law provisions, an explanation for the response can be provided.

[0098] While the above description contains many specific details, it should be construed as illustrative of preferred embodiments rather than limiting the scope of the invention. Therefore, the invention should be defined not by the described embodiments, but by the claims and their equivalents.

Claims

1. A system that provides a question-answering service using a first language model provided by an external system. An interface section that receives queries for a conversation service from a user terminal connected through an information and communications network and sends a response; A searcher including a first prompt in which a plurality of domain lists are set, and inputting a query entered through the interface section according to the first prompt into the first language model to receive a judgment result on whether a knowledge-based explanation is necessary and classifying the query as an everyday conversation or a knowledge-based conversation; and Includes an information database storing a large amount of relevant information about knowledge-based conversations, The above search engine, A conversational artificial intelligence system providing a composite question-answering service, wherein the conversation process is conducted in an end-to-end manner with the first language model for queries classified as daily conversations, and the conversation process is conducted in one or more pipeline manners by referencing the information database for queries classified as knowledge-based conversations.

2. In paragraph 1, The above search engine, A conversational artificial intelligence system providing a complex question-answering service, wherein, for a query classified as a daily conversation, a detailed query is input and a response is output, including a second prompt including multiple topics related to daily life, to the first language model, and the response is provided to the user terminal in multiple steps.

3. In paragraph 2, The above daily conversation is, An interactive artificial intelligence system that provides a complex question-answering service that includes questions and answers about the current weather and the user's schedule.

4. In paragraph 3, The above search engine, An interactive artificial intelligence system providing a composite question-answering service, wherein, when a user requests to add or change a schedule, the second prompt is updated to reflect the result of the request in the response.

5. In paragraph 1, The above search engine, A conversational artificial intelligence system providing a composite question-answering service, wherein, for a query classified as a specialized domain conversation, a second language model is used to search the information database to extract one or more related pieces of information matching the query, and a third prompt including the extracted related information is used to input a detailed query to the first language model, output a response, and provide the result to a user terminal.

6. In paragraph 5, The above second language model is, An interactive artificial intelligence system providing a complex question-answering service, which is fine-tuned by using the above detailed questions and related information corresponding to the above detailed questions as a dataset.

7. In paragraph 5, The above knowledge-based conversation is, A conversational artificial intelligence system that provides a complex question-answering service that includes legal domain-related questions and answers.

8. In paragraph 7, The above search engine, An interactive artificial intelligence system providing a composite question-answering service, wherein the third prompt is updated to proceed with the conversation process through a first pipeline that processes questions and answers regarding articles of incorporation in a machine-reading manner and a second pipeline that processes them in a legal professional advice-generating manner.

9. A method for providing a question-answering service using a first language model provided by an external system by a conversational artificial intelligence system, (a) A step of receiving a query for a conversation service from a user terminal connected through an information and communications network; (b) a step of inputting a query entered through a first prompt in which a list of multiple domains is set into the first language model; (c) a step of receiving a judgment result on whether a knowledge-based explanation is necessary from the first language model and classifying the query as a daily conversation or a knowledge-based conversation; and (d) a step of conducting a conversation process in an end-to-end manner with the first language model for a query classified as a daily conversation, or conducting a conversation process in one or more pipeline manners by referencing an information database for a query classified as a knowledge-based conversation. A method for providing a composite question-answering service including:

10. In paragraph 9, Step (d) above, (d1) For a query classified as a daily conversation, a step of inputting a detailed query through a second prompt including multiple topics related to daily life into the first language model; (d2) a step of outputting a response to the above detailed query and providing it to the user terminal; and (d3) A step of repeating steps (d1) and (d2) until the conversation ends. A method for providing a composite question-answering service including:

11. In Article 10, The above daily conversation is, A method of providing a composite question-answering service that includes questions and answers about the current weather and the user's schedule.

12. In paragraph 11, The above step (d1) is, When a user requests to add or change a schedule, the step of updating the second prompt to reflect the result of the request in the response. A method for providing a composite question-answering service including:

13. In paragraph 9, Step (d) above, (d4) For a query classified as a specialized domain conversation, a step of searching the information database using a second language model to extract one or more relevant pieces of information matching the query; (d5) a step of inputting a detailed query to the first language model through a third prompt including the extracted relevant information; (d6) a step of receiving a response from the first language model and providing it to the user terminal; and (d7) A step of repeating steps (d4) to (d6) until the conversation ends. A method for providing a composite question-answering service including:

14. In paragraph 13, Prior to step (d) above, A step of fine-tuning the second language model using the detailed query and related information corresponding to the detailed query as a dataset. A method for providing a composite question-answering service including:

15. In paragraph 13, The above knowledge-based conversation is, A method for providing a composite question-answering service including legal domain related questions and answers.

16. In paragraph 15, The above step (d4) is, Step of updating the third prompt to proceed with the conversation process by dividing it into a first pipeline that processes questions and answers about the articles of incorporation in a machine-reading manner and a second pipeline that processes them in a legal professional-advice-generating manner. A method for providing a composite question-answering service including:

Images