Dialogue system and dialogue method based on large language model and medium

Abstract

The invention discloses a dialogue system based on a large language model, a dialogue method and a medium, the system comprises a base library, a multi-agent system and a retrieval component module, the multi-agent system is provided with a plurality of subsystems, and the subsystems are used for receiving, processing and replying tasks; the base library comprises a base corresponding to each subsystem; and the retrieval component module comprises a retrieval component and a standard operation program library. According to the invention, the plurality of subsystems are arranged to process tasks together with the base library and the retrieval component module, and each subsystem can assist and complete complex tasks in an ordered operation mode, so that the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a dialogue system, a dialogue method and a medium based on a large language model. Background Art

[0002] With the development of artificial intelligence (ChatGPT), the versatility and intelligence of intelligent question-answering robots have been significantly improved, thanks to the powerful contextual semantic understanding and reasoning capabilities of large language models (LLMs).

[0003] As we all know, there are a large number of complex task communication scenarios in vertical fields (for example, insurance plan communication, house purchase plan communication, and treatment plan communication). Such scenarios are usually more complex, and the communication process must be based on rigorous domain knowledge and business logic to make real-time, controllable, and effective decision responses. In most cases, repeated communication is required before both parties can reach a feasible and satisfactory solution. For scenarios with such complexity, traditional task-based robots are almost impossible to solve effectively due to the inability to enumerate all possible dialogue scripts. Even if ChatGPT's model uses technical frameworks such as "LLMs+external knowledge" or "LLMs+cot+tools" (cot stands for Chain of Thought), it is still unable to effectively solve such complex task communication scenarios. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a large language model-based dialogue system, dialogue method and medium that can be applied to financial technology or other related fields. It is intended to jointly process tasks by setting up several subsystems with a base library and a retrieval component module. Each subsystem can assist and complete complex tasks in an orderly manner to improve work efficiency.

[0005] The technical solution of the present invention is as follows: A large language model-based dialogue system, the system comprising: a base library, a multi-agent system and a retrieval component module, wherein: the multi-agent system is provided with a plurality of subsystems; the base library comprises a plurality of bases correspondingly connected to the subsystems; the retrieval component module comprises a retrieval component and a standard operating procedure library;

[0006] Each of the subsystems is used to receive a dialog task, process the dialog task, and reply to the task according to a reply strategy matched by the retrieval component from a standard operating procedure library.

[0007] In one embodiment, the several subsystems are: an information extraction module, an information processing module and a decision response module;

[0008] The information extraction module is used to receive the dialogue information input by the user and extract key information from the dialogue information;

[0009] The information processing module is used to process the key information and generate standard information;

[0010] The decision-making reply module is used to select a reply strategy according to the standard information, and output reply information according to the reply strategy and the standard information.

[0011] In one embodiment, the intelligent model includes several subsystems, including:

[0012] Each subsystem selects a processing method to process the task according to the task type.

[0013] In one embodiment, the retrieval component is specifically configured to dynamically match the standard operating procedures in the standard operating procedure library according to the standard information.

[0014] In one embodiment, the retrieval component module includes standard operating procedures:

[0015] The standard operating procedure is used for the subsystem to process tasks according to the matching standard operating procedure.

[0016] In one embodiment, the base library matches each subsystem with a base in the base library according to characteristics of each subsystem.

[0017] In one embodiment, the information extraction module is specifically used to pre-process the dialogue information to generate text information, and extract the text information through an algorithm to obtain key information.

[0018] In one embodiment, the information processing module is specifically used to map the key information into a standard information model, process the key information according to preset classification rules, and generate standard information.

[0019] A large language model-based dialogue method, comprising:

[0020] Acquire the dialogue information input by the user, and extract key information from the dialogue information;

[0021] Processing the key information to generate standard information;

[0022] A reply strategy is matched according to the standard information, and reply information is output according to the reply strategy and the standard information.

[0023] A non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned large language model-based dialogue method.

[0024] Beneficial effects: The present invention discloses a dialogue system, device, system and medium based on a large language model. Compared with the prior art, the embodiments of the present invention are provided with several subsystems to jointly process tasks with a base library and a retrieval component module. Each subsystem can assist and complete complex tasks in an orderly manner, thereby improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0026] Figure 1 A schematic diagram of a large language model-based dialogue system provided by an embodiment of the present invention;

[0027] Figure 2 A flow chart of a large language model-based dialogue method provided by an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of the hardware structure of a large language model-based dialogue system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments of the present invention are described below in conjunction with the accompanying drawings.

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0032] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0033] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0034] See also Figure 1 , Figure 1 A schematic diagram of an embodiment of a large language model-based dialogue system provided by the present invention. The large language model-based dialogue system provided in this embodiment is applied to a system consisting of a terminal device, a network and a server, wherein the network is a medium for directly providing a communication link between the terminal device and the server, which may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.; the operating system on the terminal device may include a handheld device operating system (iPhone operating system, iOS system), Android system or other operating system, and the terminal device is connected to the server through the network to achieve interaction, thereby performing operations such as receiving or sending data, etc., and specifically may be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, portable computers and desktop servers, etc.

[0035] like Figure 1 As shown, Figure 1 A large language model-based dialogue system is provided in an embodiment of the present invention, and the system specifically includes:

[0036] The system includes: a base library 13, a multi-agent system 11 and a retrieval component module 12, wherein: the multi-agent system has several subsystems set according to the task, and the subsystems are used to receive, process and reply tasks; the base library includes a base corresponding to each subsystem; the retrieval component module includes a retrieval component and a standard operating procedure library. Each of the subsystems is used to receive a dialogue task, process the dialogue task and reply to the task according to the reply strategy matched by the retrieval component from the standard operating procedure library, thereby outputting reply information.

[0037] In this embodiment, a multi-agent dialogue system (MAS) based on large language models (LLMs) is proposed. It allows multiple agents (subsystems) to communicate and collaborate with each other in a dialogue to achieve a common goal or task. Agents can share information and knowledge to collaboratively solve problems or complete tasks. Each agent can independently control its own behavior and internal state, and agents can communicate and exchange information. Agents can understand and follow social rules and protocols. The system can dynamically adjust its response strategy based on the context of the dialogue and the behavior of the agent. Each agent can have a different knowledge background, personality or role, making the dialogue richer and more interesting. The agent can understand the user's input and provide corresponding feedback to form an effective human-computer interaction. The agent can understand the context of the dialogue, including the previous dialogue history and the current topic of discussion.

[0038] Specifically, complex problems are broken down into multiple independent modules, and an Agent based on LLMs collaborates to complete the above complex solutions for communication. Multi-Agent Systems (MAS) are composed of multiple sub-agents (sub-agents), (the number of Agents and tasks are designed as needed according to the complexity of the task), each sub-agent has a certain degree of autonomy and ability, can perform tasks independently or assume a specific role in the system, and each sub-agent assists and completes the entire complex task in an orderly manner. Each sub-agent can control its own behavior and decision-making process and is not directly controlled by other agents. Sub-agents can communicate and exchange information to coordinate their actions and reach consensus. In a multi-agent system, a task can be assigned to the sub-agent that is most suitable to complete it. Furthermore, in addition to the sub-agent, the Multi-Agent framework introduces other functional components as needed to coordinate the sub-agents to jointly complete the complex solution communication task. The base library includes the base corresponding to each subsystem. The base library is the LLMs library, which generally refers to the universal base, because each agent must use the base. The large language model is the basic ability of these agents. Some agents may need a large language model to use the base, while others may only need a small model. The retrieval component module includes a retrieval component (such as an SOP retrieval component) and a standard operating procedure library (such as an SOP library), which is used for the retrieval component to dynamically match the corresponding standard operating procedures in the standard operating procedure library according to the multi-agent system. Based on the powerful understanding and reasoning ability of LLMs, each sub-agent has its own responsibilities, which can focus more on solving sub-module tasks and effectively ensure the effect of sub-task solving.

[0039] Furthermore, for tasks that require a large amount of external knowledge support, it is necessary to integrate and update the knowledge base. By designing a structured knowledge base, it is possible to store, retrieve, and update knowledge of different types and fields. Determine the source of knowledge, including public databases, professional websites, academic papers, industry reports, etc. Collect data from different sources using crawlers, API calls, or data import tools. Clean the collected data to remove errors, duplications, and irrelevant information. Select appropriate knowledge representation methods, such as ontology, graph databases, semantic networks, etc. Integrate knowledge from different sources and resolve conflicts and inconsistencies. Classify and annotate knowledge for easy retrieval and use. Verify the correctness and reliability of the integrated knowledge. Design an automated update mechanism to regularly check and update the knowledge base content. Develop a query interface so that the agent can easily retrieve the knowledge base. Apply reasoning rules to extract implicit information and associations from the knowledge base. Regularly maintain the knowledge base, including data backup, performance optimization, etc. Integrate the knowledge base with the agent to ensure that the agent can access and apply the knowledge base according to task requirements. Improve the interpretability of knowledge base operations to help understand the agent's decision-making process. Through these steps, the effective integration and continuous updating of the knowledge base can be ensured, providing a solid foundation for tasks that require a large amount of external knowledge support.

[0040] In one embodiment, the multi-agent system includes several subsystems, including: an information extraction module, an information processing module and a decision response module, wherein the information extraction module is used to receive dialogue information input by the user and extract key information from the dialogue information; the information processing module is used to process the key information and generate standard information; the decision response module is used to select a response strategy according to the standard information and output response information according to the response strategy and the standard information.

[0041] In this embodiment, the multi-agent system includes several subsystems, and each sub-agent has different functions, for example, including an information extraction module, an information processing module, and a decision response module. The information extraction module Agent 1 is responsible for receiving the input of the user's dialogue information, preprocessing the input dialogue information, and using named entity recognition (NER) to identify specific entities. Use TF-IDF or TextRank and other technologies to extract useful information from the dialogue information to generate key information. Identify user intentions and emotional tendencies. Output the extracted information for the next agent to use. The information processing module Agent 2 is responsible for further processing the extracted key information, converting it into standardized and structured data, and generating standard information for subsequent processing. Receive the output of the information extraction agent. Map and convert the extracted information, such as converting non-standard occupation names into standard occupation names. Classify and label information, such as classifying occupations into specific occupational categories. Verify and format information to ensure the accuracy and consistency of information. Output the processed information for the decision response agent to use.

[0042] For example, in the underwriting scenario, the occupation name "coder" in the user's query is extracted and processed into the standard information required for decision-making, "occupation name - computer development; occupation category - category 1 occupation", so that the decision-making response module Agent3 can process the response. The decision-making response module Agent3 dynamically selects the decision SOP corresponding to the current information based on the standard information, and then allows the large language model to make accurate decision responses based on the SOP and standard information. Receive the output of the information processing agent. Use the retrieval component to dynamically match the corresponding SOP in the SOP library based on the processed information. Apply the matched SOP to guide the large language model to make decisions. Generate a decision response and ensure the accuracy and relevance of the response. Send the decision response to the user or other parts of the system. This process ensures that each step from the original conversation to the final response is efficient and accurate, and each agent can focus on its specific task. Through this division of labor and cooperation, the multi-agent system can better handle complex tasks and provide high-quality services.

[0043] In one embodiment, the intelligent model includes several subsystems, including:

[0044] Each subsystem selects a processing method to process the task according to the task type.

[0045] In this embodiment, each LLMs-based sub-agent can be implemented by the method of Prompt+LLMs+external knowledge base, or by the simple Prompt+LLMs method to implement the sub-model task, independently complete the current module task, and output the result to the next Agent. Different Agents decide whether to use only "Prompt+LLMs" or "prompt+LLMS+external knowledge base" according to the characteristics and complexity of the processing task. For example: Information extraction Agent: The task is relatively simple and does not need to rely on external knowledge. Information extraction can be achieved through Prompt+LLMs. The "Prompt+LLMs" method is adopted to use carefully designed prompts (Prompt) to guide large language models (LLMs) for information extraction. It is suitable for scenarios where the task is relatively simple and the model itself can handle it. For example: Information processing and decision response Agent tasks are relatively complex, and processing needs to rely on external knowledge (occupational category table, decision SOP) to achieve information processing and decision response. The "Prompt+LLMs+external knowledge base" method is required to combine the external knowledge base to enhance the decision-making ability of the model. Suitable for scenarios where the task is complex and needs to rely on external data or rules for information processing. For example, it may be necessary to combine "Prompt+LLMs+external knowledge base" to ensure the accuracy and reliability of decision-making. Suitable for scenarios where it is necessary to comprehensively consider the current information and the SOP in the external knowledge base to formulate a response.

[0046] Specifically, the subsystem first needs to identify the type of task received, which may be based on the task description, user input, or preset task classification. Based on the identified task type, the system needs to route the task to the appropriate subsystem or processing module. Manage the task queue to ensure that tasks are correctly queued for processing according to priority and type. Define processing strategies or algorithms for each task type and ensure that the subsystem can select the corresponding strategy based on the task type. Dynamically allocate computing resources and storage resources according to task requirements. Ensure that the subsystem is designed to be modular so that different processing modules can be quickly adjusted and combined for different task types. The subsystem needs to understand the context information of the task so that the processing method can be selected more accurately. Use the rule engine to process tasks based on predefined rules, such as decisions based on IF-THEN rules. For tasks that require pattern recognition or prediction, use trained machine learning models for processing. Execute the selected processing method, which may include data retrieval, information extraction, decision making, etc. After the task processing is completed, verify the correctness and completeness of the results. Perform iterative optimization based on the execution results and user feedback to improve the processing method. Monitor the performance of the subsystem in processing tasks to ensure that the service level agreement (SLA) is met. In this way, each subsystem in the multi-agent system can flexibly choose the most appropriate processing method according to the task type, thereby improving the efficiency and effectiveness of the entire system.

[0047] Furthermore, set the priority of the task according to the urgency, importance or other business rules. Dynamically assign tasks according to the current load and processing capacity of the Agent. Use load balancing technology to distribute tasks and avoid overloading some Agents. Monitor the length and processing speed of the task queue in real time to adjust resources in time. Use efficient task scheduling algorithms such as shortest job first (SJF) and round-robin scheduling. Manage the dependencies between tasks to ensure that tasks are executed in the correct order. Allocate appropriate system resources according to the type and complexity of the task. Assign different types of tasks to different queues to improve processing efficiency. With this flexible strategy, the multi-agent system can choose the most appropriate processing method based on the specific needs of each Agent and the complexity of the task.

[0048] In one embodiment, the retrieval component module includes a retrieval component for dynamically matching the standard operating procedures in the standard operating procedure library according to the standard information.

[0049] In this embodiment, the retrieval component refers to the field processed according to the information, dynamically matching the corresponding SOP in the SOP library, and the retrieval function. In a multi-agent system, the role of the retrieval component is to dynamically match the corresponding standard operating procedures in the standard operating procedure library (SOP library) according to the standard information. The format and content of the standard information are clarified to ensure the consistency and matchability of the information. The SOP library is indexed to speed up the retrieval speed, and the index can be based on task type, keywords, topics, etc. The retrieval component receives standard information from the information processing agent. The received standard information is preprocessed, such as format unification, keyword extraction, etc. Define a retrieval strategy, including a retrieval algorithm (such as full-text search, keyword matching, etc.) and a sorting logic. According to the preprocessed standard information, a dynamic match is performed in the SOP library using a retrieval strategy. Evaluate the relevance and accuracy of the matching results, and it may be necessary to apply a machine learning model for scoring. According to the evaluation results, the matching SOPs are sorted and the most appropriate operating procedures are selected. Consider the context information of the current task and optimize the matching results. Regularly update the SOP library and retrieval components to adapt to new task requirements and best practices. The retrieval component is integrated into the multi-agent system for comprehensive testing to ensure that it works in conjunction with other components. Through these steps, the retrieval component can efficiently match the most appropriate standard operating procedures based on standard information and provide accurate guidance for the decision-making response agent, thereby improving the quality and efficiency of the entire system in processing tasks.

[0050] Furthermore, in order to ensure that the retrieval component can accurately match SOPs in different task types, clarify the characteristics and classification standards of different tasks, and establish a clear description for each task type. Ensure that the SOP library contains detailed operating procedures for various task types and keep the content updated. Implement advanced retrieval algorithms such as fuzzy matching, semantic search, or machine learning-driven retrieval. Enable the retrieval component to understand the task context so that it can better match applicable SOPs. Perform a multi-dimensional evaluation of the matching results, including task urgency, user satisfaction, etc., to ensure that the matched SOP is the most appropriate. The retrieval component can adapt to different task types, achieve more accurate SOP matching, and improve the overall performance and user satisfaction of the multi-agent system.

[0051] In one embodiment, the retrieval component module includes a standard operating procedure for the subsystem to process the task according to the matched standard operating procedure.

[0052] In this embodiment, SOP usually refers to "Standard Operating Procedure" (SOP library), which is a set of detailed, written instructions used to guide the execution of specific tasks or activities. SOP is a key component in a multi-agent system. It provides necessary guidance and specifications for subsystems to ensure that tasks can be performed in accordance with established best practices and standards. Formulating and following SOPs helps reduce errors and risks and improve overall work performance. Collect existing standard operating procedures or create new SOPs based on business needs. Classify SOPs according to standards such as task type, field, and complexity. Digitize SOP content so that it can be indexed and retrieved by retrieval components. Ensure that all SOPs follow a unified format and standard for easy understanding and application. Index the SOP library to improve retrieval efficiency and accuracy. Develop a retrieval interface that allows subsystems to retrieve corresponding SOPs based on task characteristics. Version control changes to SOPs, retain historical records, and support version backtracking. Ensure that the SOP library can be integrated with other components in the multi-agent system (such as retrieval components and decision support systems). Through these steps, the SOP library is able to provide accurate guidance to the subsystems in the multi-agent system, ensuring that tasks are handled efficiently and in compliance with standards, while improving the reliability and consistency of the entire system.

[0053] Furthermore, define the core elements and structure of the SOP to ensure that all operating procedures contain key information such as purpose, scope, steps, responsibilities, etc. Break down the SOP into modular components, allowing these components to be combined or modified according to different business scenarios. Provide standardized templates and frameworks to allow a certain degree of customization while maintaining consistency. Include flexibility options in the SOP, such as conditional statements, optional steps, or alternative methods, to adapt to different business needs. Allow specific parameters to be entered when executing the SOP to adjust the operating procedures to suit different tasks or environments. Enable the SOP to dynamically adjust its content based on contextual information (such as task type, user role, environmental conditions). Ensure that the SOP library has both the necessary standardization and the flexibility to adapt to different business scenarios to improve its practicality and effectiveness.

[0054] In one embodiment, the base library matches the bases in the base library according to the characteristics of each subsystem.

[0055] In this embodiment, in a multi-agent system, the base library (Base Model Library) usually contains a series of pre-trained large language models (LLMs), which serve as the basic capabilities of the subsystems. In order to effectively match the bases in the base library according to the characteristics of each subsystem. Analyze the functions, task types, performance requirements and resource constraints of each subsystem in detail. Evaluate the performance, characteristics and applicable scenarios of each model in the base library. Match the requirements of the subsystem with the characteristics of the base model and select the most suitable model. Consider the computing resources of the subsystem (such as CPU, GPU, memory), and select a base model whose resource consumption matches the performance. Perform performance tests on candidate base models to ensure that they meet the performance requirements of the subsystem. Ensure that each subsystem can match its most suitable base model, thereby improving the performance and efficiency of the entire multi-agent system.

[0056] Furthermore, in order to ensure that the models in the base library can meet the specific needs of different subsystems, careful planning, selection, customization and maintenance are required. Conduct in-depth analysis of the functions, performance requirements, resource limitations and usage scenarios of each subsystem. Evaluate the performance indicators, advantages, limitations and applicable scenarios of each model in the base library. Match the models in the base library according to the characteristics of the subsystem and select the model that best meets the needs. A modular design is adopted to enable the base model to be flexibly integrated with other system components. Allow customized adjustments or fine-tuning of the base model to better meet the needs of specific subsystems. Consider the computing resources of the subsystem and select or adjust the model to optimize resource utilization efficiency. Perform performance tests on the selected model to ensure that it meets the performance standards of the subsystem. It can ensure that the models in the base library meet current needs.

[0057] In this embodiment, in one embodiment, the information extraction module is used to receive dialogue information input by the user and extract key information from the dialogue information, which includes: preprocessing the dialogue information to generate text information, extracting the text information through an algorithm to obtain key information.

[0058] In this embodiment, the information extraction module is a key component in the multi-agent system, which is responsible for receiving the dialogue information input by the user and extracting key information from it for further processing by the system. Specifically, the original conversation input of the user is received, which may include text, voice or other forms of data. The dialogue information input by the user is pre-processed for cleaning and standardization, including removing noise, removing irrelevant characters, punctuation marks, converting to a unified text format, etc. The NER (Named Entity Recognition) technology is used to identify entities in the text. Common NER entity categories include PER (name), LOC (location), ORG (organization), DATE (date), TIME (time), MONEY (currency), NUM (numerical value), etc. The key information in the text is extracted using algorithms, such as TF-IDF (Term Frequency-Inverse Document Frequency), TextRank (text summary algorithm), etc. Dependency syntactic analysis is applied to determine the dependency relationship between words in the text, which helps to understand the sentence structure. Identify the intention behind the user input, which can be achieved through pattern matching or machine learning models. Evaluate the sentiment of user input, such as positive, negative, or neutral. Identify actions (predicates) in sentences and roles associated with the actions (such as agents, patients). Identify relationships between entities in text, such as the "founder" relationship between "Apple" and "Steve Jobs". Identify events and their participants in text, such as "signer" and "contract" in the "signing contract" event. Extract core information from long texts and generate summaries. Consider the conversation history and current context to extract information more accurately. If the input includes multiple modalities such as images and sounds, the corresponding technology needs to be used to extract key information from each modality. Depending on the specific application scenario, it may be necessary to define some custom rules to extract specific information. Integrate the information extraction agent into the entire system and conduct comprehensive testing to ensure its accuracy and reliability. The information extraction agent can effectively extract key information from conversation information.

[0059] In one embodiment, the information processing module is used to process the key information to generate standard information, which includes: mapping the key information to a standard information model, processing the key information according to preset classification rules, and generating standard information.

[0060] In this embodiment, the information extraction module receives key information extracted from the user input query, such as the occupation name "coder". Design a standard information model to clarify the format and content of the information required for decision-making, such as "occupation name" and "occupation category". Map the key information into the standard information model. For example, map "coder" to "computer development". According to the preset classification rules, classify the occupation name into the corresponding occupation category, such as classifying "computer development" as "category 1 occupation", and generate standard information. Verify whether the processed information meets the standards and format requirements to ensure the accuracy of the information. Consider the contextual information entered by the user to ensure that the converted information is still applicable in the specific context. Use the ontology knowledge base to standardize and classify the occupation names to ensure the consistency and accuracy of the information. Apply the rule engine to handle complex information conversion logic, such as synonym processing of specific occupation names. Train the machine learning model to identify and convert information that is not easy to process through rules. Enrich the processed information and add information that may be missing but useful for decision-making. Format the processed information into a standard format that the decision-making reply agent can understand. Design a feedback mechanism to optimize the information processing process based on the feedback of the decision-making reply agent. Enable the information processing agent to learn from new data and feedback, and continuously optimize the accuracy and efficiency of information processing. Monitor the performance of the information processing agent to ensure that it can process information in a timely and accurate manner. Integrate the information processing agent into the entire system to ensure seamless connection with the information extraction module and decision response agent.

[0061] Furthermore, when the information processing agent processes key information in different languages, it needs to ensure the accuracy and consistency of the information. Specifically, ensure that the information processing agent has the ability to process multiple languages, including word segmentation, part-of-speech tagging, named entity recognition, etc. Automatically detect the language of user input so that the appropriate processing flow and resources can be selected. Build or integrate professional dictionaries and ontology knowledge bases for each language for information standardization and synonym processing. For some languages, it may be necessary to use machine translation technology to translate key information into a language that the information processing agent can process. Use cross-language embedding technology, such as multilingual Bert (mBERT) or XLM-R, to understand and process semantic information in different languages. Perform quality control checks on the processed information to ensure that it meets the expected standards and formats. Consider contextual information when processing information to ensure the consistency of information in different language environments. Perform multilingual testing on the information processing agent to ensure performance and accuracy in different language environments. Convert information processed in different languages ​​into a unified standardized format for subsequent processing. The information processing agent is able to process key information in different languages ​​while ensuring the accuracy and consistency of the information.

[0062] A non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned large language model-based dialogue method.

[0063] like Figure 2 As shown, Figure 2 This is a flow chart of a large language model-based dialogue method provided by an embodiment of the present invention. The method specifically includes the following steps:

[0064] S100, obtaining dialogue information input by a user, and extracting key information from the dialogue information;

[0065] S200, processing the key information to generate standard information;

[0066] S300: Match a reply strategy according to the standard information, and output reply information according to the reply strategy and the standard information.

[0067] In this embodiment, a user interface is designed to allow users to submit dialogue information in text, voice or other forms. The user input is cleaned and standardized, including removing noise, punctuation, converting to uniform uppercase and lowercase, etc. Information extraction technology (such as NER, keyword extraction, and intent recognition) is used to extract key information from user input. The extracted key information is converted into a standardized format for subsequent processing and analysis. The key information is further processed and verified to ensure the accuracy and completeness of the information. According to the standard information, the most appropriate reply strategy or SOP is retrieved and matched. The reply strategy that best matches the current context and user intent is selected from the retrieved strategies. The selected reply strategy is used to guide the generation of reply information. Combine the standard information and the reply strategy to generate reply information or perform corresponding operations. The multi-agent system can effectively process the dialogue information input by the user and generate accurate, relevant and user-friendly reply information.

[0068] Another embodiment of the present invention provides a hardware system based on a large language model dialogue, such as Figure 3 As shown, the system 10 includes:

[0069] One or more processors 110 and memory 120, Figure 3 A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0070] The processor 110 is used to complete various control logics of the system 10, and it can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0071] The memory 120 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the large language model-based dialogue system in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the system 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, realizing the large language model-based dialogue system in the above embodiment.

[0072] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the system 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the system 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0073] One or more units are stored in the memory 120, and when executed by one or more processors 110, the following procedures are implemented:

[0074] The large language model-based dialogue system includes: a base library, a multi-agent system and a retrieval component module, wherein: the multi-agent system is provided with a plurality of subsystems, and the subsystems are used to receive, process and reply tasks; the base library includes a base corresponding to each subsystem; the retrieval component module includes a retrieval component and a standard operating procedure library.

[0075] In one embodiment, the multi-agent system includes several subsystems, including: an information extraction module, an information processing module and a decision response module, wherein the information extraction module is used to receive dialogue information input by the user and extract key information from the dialogue information; the information processing module is used to process the key information and generate standard information; the decision response module is used to select a response strategy according to the standard information and output response information according to the response strategy and the standard information.

[0076] In one embodiment, the intelligent model includes several subsystems, including:

[0077] Each subsystem selects a processing method to process the task according to the task type.

[0078] In one embodiment, the retrieval component module includes a retrieval component, including:

[0079] The retrieval component is used to dynamically match the standard operating procedures in the standard operating procedure library according to the standard information.

[0080] In one embodiment, the retrieval component module includes a standard operating procedure, including:

[0081] The standard operating procedure library is used for the subsystem to process tasks according to matching standard operating procedures.

[0082] In one embodiment, it includes:

[0083] The base library matches the bases in the base library according to the characteristics of each subsystem.

[0084] In one embodiment, the information extraction module is used to receive the dialogue information input by the user and extract key information from the dialogue information, including:

[0085] The dialogue information is preprocessed to generate text information, and the text information is extracted through an algorithm to obtain key information.

[0086] In one embodiment, the information processing module is used to process the key information to generate standard information, including:

[0087] The key information is mapped into a standard information model, and the key information is processed according to preset classification rules to generate standard information.

[0088] A large language model-based dialogue method, comprising:

[0089] Acquire the dialogue information input by the user, and extract key information from the dialogue information;

[0090] Processing the key information to generate standard information;

[0091] A reply strategy is matched according to the standard information, and reply information is output according to the reply strategy and the standard information.

[0092] An embodiment of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions, which are executed by one or more processors, wherein the computer program implements the following when executed by the processor: the system includes: a base library, a multi-agent system and a retrieval component module, wherein: the multi-agent system is provided with a plurality of subsystems, and the subsystems are used to receive, process and reply to tasks; the base library includes a base corresponding to each subsystem; the retrieval component module includes a retrieval component and a standard operating procedure library.

[0093] In one embodiment, the multi-agent system includes several subsystems, including: an information extraction module, an information processing module and a decision response module, wherein the information extraction module is used to receive dialogue information input by the user and extract key information from the dialogue information; the information processing module is used to process the key information and generate standard information; the decision response module is used to select a response strategy according to the standard information and output response information according to the response strategy and the standard information.

[0094] In one embodiment, the intelligent model includes several subsystems, including:

[0095] Each subsystem selects a processing method to process the task according to the task type.

[0096] In one embodiment, the retrieval component module includes a retrieval component, including:

[0097] The retrieval component is used to dynamically match the standard operating procedures in the standard operating procedure library according to the standard information.

[0098] In one embodiment, the retrieval component module includes a standard operating procedure, including:

[0099] The standard operating procedure library is used for the subsystem to process tasks according to matching standard operating procedures.

[0100] In one embodiment, it includes:

[0101] The base library matches the bases in the base library according to the characteristics of each subsystem.

[0102] In one embodiment, the information extraction module is used to receive the dialogue information input by the user and extract key information from the dialogue information, including:

[0103] The dialogue information is preprocessed to generate text information, and the text information is extracted through an algorithm to obtain key information.

[0104] In one embodiment, the information processing module is used to process the key information to generate standard information, including:

[0105] The key information is mapped into a standard information model, and the key information is processed according to preset classification rules to generate standard information.

[0106] A large language model-based dialogue method, comprising:

[0107] Acquire the dialogue information input by the user, and extract key information from the dialogue information;

[0108] Processing the key information to generate standard information;

[0109] A reply strategy is matched according to the standard information, and reply information is output according to the reply strategy and the standard information.

[0110] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0111] In summary, the present invention discloses a large language model-based dialogue system, dialogue method and medium, the system includes: a base library, a multi-agent system and a retrieval component module, wherein: the multi-agent system is provided with a plurality of subsystems, the subsystems are used to receive, process and reply tasks; the base library includes a base corresponding to each subsystem; the retrieval component module includes a retrieval component and a standard operating procedure library. The present invention is provided with a plurality of subsystems to jointly process tasks with the base library and the retrieval component module, each subsystem can assist and complete complex tasks in an orderly manner, thereby improving work efficiency.

[0112] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the above-described equipment, devices and units can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. Those of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0113] In the several embodiments provided by the present invention, it should be understood that the disclosed equipment, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. Units with the same function may also be combined into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices or units, or may be an electrical, mechanical or other form of connection.

[0114] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0115] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0116] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a computer-readable storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the specification described in each embodiment of the present invention. The aforementioned computer-readable storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a magnetic disk or an optical disk.

[0117] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, a floppy disk, a flash memory, an optical storage device, etc.

[0118] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A dialogue system based on a large language model, It is characterized in that include: A base library, a multi-agent system and a retrieval component module, wherein: the multi-agent system is provided with a plurality of subsystems; the base library includes a plurality of bases correspondingly connected to the subsystems; the retrieval component module includes a retrieval component and a standard operating procedure library; Each of the subsystems is used to receive a dialog task, process the dialog task, and reply to the task according to a reply strategy matched by the retrieval component from a standard operating procedure library.

2. The large language model-based dialogue system according to claim 1, It is characterized in that The several subsystems are: information extraction module, information processing module and decision response module; The information extraction module is used to receive the dialogue information input by the user and extract key information from the dialogue information; The information processing module is used to process the key information and generate standard information; The decision-making reply module is used to select a reply strategy according to the standard information, and output reply information according to the reply strategy and the standard information.

3. The large language model-based dialogue system according to claim 2, It is characterized in that Each subsystem selects a processing method to process the task according to the task type.

4. The large language model-based dialogue system according to claim 3, It is characterized in that The retrieval component is specifically used to dynamically match the standard operating procedures in the standard operating procedure library according to the standard information.

5. The large language model-based dialogue system according to claim 4, It is characterized in that The retrieval component module includes standard operating procedures: The standard operating procedure is used for the subsystem to process tasks according to the matching standard operating procedure.

6. The large language model-based dialogue system according to claim 5, It is characterized in that The base library matches each subsystem with the base in the base library according to the characteristics of each subsystem.

7. The large language model-based dialogue system according to claim 2, It is characterized in that The information extraction module is specifically used to pre-process the dialogue information to generate text information, and extract the text information through an algorithm to obtain key information.

8. The large language model-based dialogue system according to claim 2, It is characterized in that The information processing module is specifically used to map the key information into a standard information model, process the key information according to preset classification rules, and generate standard information.

9. A large language model-based dialogue method according to claim 1, It is characterized in that include: Acquire conversation information, and extract key information from the conversation information; Processing the key information to generate standard information; A reply strategy is matched according to the standard information, and reply information is output according to the reply strategy and the standard information.

10. A non-volatile computer-readable storage medium, It is characterized in that The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by one or more processors, enable the one or more processors to execute the large language model-based dialogue method according to claim 9.

Images