Multi-agent collaborative intelligent reply method and device, equipment and medium

Abstract

The invention provides a multi-agent collaborative intelligent reply method and device, equipment and a medium, relates to the technical field of artificial intelligence, and is suitable for the financial field and the medical field. The method comprises the following steps: performing intention recognition on a target problem through an intention recognition agent to obtain a target intention; performing concept extraction on the target question through a concept extraction agent to obtain a target concept, the target concept including the first new knowledge; updating the first new knowledge to the initial knowledge graph through the dynamic updating agent to obtain a first knowledge graph; task planning is conducted on the target intention and the target concept through a task planning agent, a task sequence is obtained, and the task sequence comprises a target task; performing data reading on the first knowledge graph through the graph interaction agent and the target task to obtain target sub-graph data; and performing answer generation on the target sub-map data and the target question through the response generation agent to obtain a target answer. According to the invention, the accuracy of intelligent reply is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and is applicable to the financial and medical fields. In particular, it relates to a method, device, or medium for intelligent response involving multiple agents. Background Technology

[0002] An agent is an autonomous entity capable of perceiving, making decisions, and taking actions to achieve a goal within a specific environment. For example, in customer service scenarios in fintech, if a customer asks, "Where can I download my transaction statements for the past month?", an agent can answer this question and provide steps to download the statements. Similarly, in patient self-help Q&A scenarios in the healthcare field, if a patient asks, "What are the early symptoms of high blood sugar?", an agent can process this question and provide health advice to the patient in simple, easy-to-understand language.

[0003] Intelligent agents using related technologies can provide intelligent responses using knowledge graphs. However, these technologies typically treat knowledge graphs as static information sources, but user-input questions may involve new knowledge not found in the knowledge graph, leading to lower accuracy in intelligent responses. Summary of the Invention

[0004] The main objective of this application is to propose a multi-agent collaborative intelligent response method, apparatus, device, and medium, which can solve the technical problem that the accuracy of intelligent responses is low due to the lack of knowledge related to the current user's question in the knowledge graph, thereby improving the accuracy of intelligent responses.

[0005] To achieve the above objectives, a first aspect of this application proposes a multi-agent collaborative intelligent response method, the method comprising: Identify the target problem; The target intent is obtained by identifying the intent of the target question through an intent-recognition intelligent agent. A concept extraction agent extracts concepts from the target problem to obtain target concepts; wherein the target concepts include first new knowledge. The first knowledge graph is obtained by dynamically updating the first new knowledge to the preset initial knowledge graph through the intelligent agent; A task planning agent performs task planning on the target intent and the target concept to obtain a task sequence, which includes the target task. The target sub-graph data is obtained by reading data from the first knowledge graph through the graph interaction agent and the target task. The response-generating agent generates an answer to the target sub-map data and the target question, thus obtaining the target answer.

[0006] Optionally, the step of updating the first new knowledge to a preset initial knowledge graph through a dynamically updating agent to obtain the first knowledge graph includes: Obtain the number of the target concepts to get the number of concepts; The new entity is determined based on each of the target concepts, including the first new knowledge; If the number of concepts is greater than or equal to 2, then the first entity relationship is extracted from the target problem based on any two newly added entities through the dynamically updated agent. All the newly added entities and the relationships between the first entities are inserted into the initial knowledge graph to obtain the first knowledge graph.

[0007] Optionally, the step of updating the first new knowledge to a preset initial knowledge graph through a dynamically updating agent to obtain the first knowledge graph further includes: If the number of concepts is equal to 1, the dynamic update agent retrieves entities in the initial knowledge graph based on the target concept to obtain the target associated entity and determines the second entity relationship between the newly added entity and the target associated entity. The newly added entity and the relationship between the second entity are inserted into the initial knowledge graph to obtain the first knowledge graph.

[0008] Optionally, the step of reading data from the first knowledge graph through a graph interaction agent and the target task to obtain target sub-graph data includes: The graph interaction agent generates a statement based on the target concept and the entities in the first knowledge graph to obtain the shortest path matching statement. The first knowledge graph is read based on the shortest path matching statement to obtain the target subgraph data.

[0009] Optionally, before generating an answer to the target sub-map data and the target question through a response-generating agent to obtain the target answer, the method further includes: The target sub-graph data is logically extracted by the reasoning agent to obtain the target logical chain data; The reasoning agent performs task induction based on the target logical chain data and the task sequence to obtain task reasoning data. The task reasoning data is used as the target sub-map data.

[0010] Optionally, the step of logically extracting the target sub-graph data through the reasoning agent to obtain the target logical chain data includes: Extract the jump paths between entities from the target sub-map data; Based on the large language model, information is filled in for the concepts and relationships in each jump path to obtain the target optimized path; Based on all the target optimization paths, path integration is performed to obtain the target logical chain data.

[0011] Optionally, after obtaining task inference data by performing task induction through the inference agent based on the target logical chain data and the task sequence, the method further includes: The concept extraction agent extracts concepts from the task reasoning data to obtain reasoning concepts; wherein the reasoning concepts include second new knowledge. The dynamically updating agent updates the first knowledge graph based on the second new knowledge to obtain the second knowledge graph.

[0012] To achieve the above objectives, a second aspect of this application proposes a multi-agent collaborative intelligent response device, the device comprising: The problem retrieval module is used to retrieve the target problem; The intent recognition module is used to recognize the intent of the target question through an intent recognition agent to obtain the target intent; A concept extraction module is used to extract concepts from the target problem through a concept extraction agent to obtain target concepts; wherein the target concepts include first new knowledge. The graph update module is used to update the first new knowledge to a preset initial knowledge graph through a dynamic updating agent, thereby obtaining the first knowledge graph; The task planning module is used to perform task planning on the target intent and the target concept through a task planning agent to obtain a task sequence, wherein the task sequence includes the target task. The graph interaction module is used to read data from the first knowledge graph through the graph interaction agent and the target task to obtain target sub-graph data; The response generation module is used to generate an answer to the target sub-map data and the target question through a response generation agent, thereby obtaining the target answer.

[0013] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the multi-agent collaborative intelligent response method described in the first aspect.

[0014] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-agent collaborative intelligent response method described in the first aspect.

[0015] The multi-agent collaborative intelligent response method, device, electronic device, and computer-readable storage medium proposed in this application operate on a multi-agent collaborative architecture comprising an intent recognition agent, a concept extraction agent, a dynamic update agent, a task planning agent, a graph interaction agent, and a response generation agent. After obtaining the target question, the intent recognition agent outputs the target intent, the concept extraction agent outputs the target concept, the dynamic update agent updates the knowledge graph, the task planning agent outputs the target task, the graph interaction agent outputs the target sub-graph data, and finally the response generation agent outputs the target answer. Since the knowledge graph is updated by the dynamic update agent before querying the graph, it eliminates the need for manual expertise to declare new knowledge and obtains question-related sub-graph data from the knowledge graph. Thus, this application, by unifying and coordinating multiple agents, forms a multi-agent collaborative architecture capable of dynamically updating the knowledge graph to achieve intelligent response, thereby improving the accuracy of intelligent responses. Attached Figure Description

[0016] Figure 1 This is a flowchart of the intelligent response method for multi-agent collaboration provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart for step 104 in the document; Figure 3 yes Figure 1 Another flowchart of step 104 in the process; Figure 4 yes Figure 1 The flowchart for step 106 in the document; Figure 5 This is a flowchart of a multi-agent collaborative intelligent response method provided in another embodiment of this application; Figure 6 yes Figure 5 The flowchart for step 501 in the document; Figure 7 This is a block diagram of the module structure of the intelligent response device for multi-agent collaboration provided in the embodiments of this application; Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] First, let's analyze some of the terms used in this application: Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0021] Natural Language Processing (NLP): NLP uses computers to process, understand, and utilize human language (such as Chinese and English). It is a branch of artificial intelligence and an interdisciplinary field of computer science and linguistics, often referred to as computational linguistics. NLP includes syntactic analysis, semantic analysis, and discourse understanding. It is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information and image processing, information extraction and filtering, text classification and clustering, sentiment analysis, and opinion mining. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computation.

[0022] An AI Agent is an application that can autonomously plan and invoke external tools to execute tasks based on objectives and external information, and reflect on the results. AI Agents are designed around task objectives and business processes in terms of implementation, operational form, and behavioral characteristics.

[0023] Large Language Models (LLMs) are a class of deep learning models that utilize massive amounts of text data for autoregressive or self-supervised learning. They are capable of generating, understanding, translating, and summarizing complex natural language tasks. Common underlying frameworks are often based on the Transformer architecture, with core elements including attention mechanisms, multi-layer stacking, and positional encoding. Typical capabilities include text generation, question answering, dialogue, summarization, translation, code completion, and sentiment analysis.

[0024] A knowledge graph (KG) is a graph-based method for representing and managing knowledge, where "entities" (nodes) are connected by "relationships" (edges) to form a queryable and reasonable knowledge network. Entities are typically real-world things or concepts (such as people, places, companies, products, events, etc.), while relationships describe the semantic connections between entities (such as "belongs to," "creator," "located in," "same category," etc.). The goal of a knowledge graph is to integrate structured, semi-structured, and unstructured data into a unified, reasonable knowledge resource, supporting semantic queries, reasoning, and the discovery of new relationships.

[0025] Currently, various attempts have been made in the industry to combine Large Language Models (LLMs) with Knowledge Graphs (KGs) to improve the accuracy and interpretability of question-answering systems. However, most existing systems suffer from several significant problems: First, most products only use knowledge graphs as static information sources, lacking the ability for users to interact with and dynamically update the graph in real time; second, complex query languages ​​(such as SPARQL or Cypher) are not user-friendly for non-technical users, greatly limiting their widespread application in professional fields such as medicine and law; third, existing systems lack modularity and process transparency, making it difficult to meet practical needs such as privacy protection, compliance tracking, and multi-step reasoning. Furthermore, most current LLM-driven question-answering systems cannot achieve a complete closed loop from user intent recognition to graph manipulation during task execution, resulting in low success rates and an inability to audit and debug intermediate reasoning processes.

[0026] Based on this, embodiments of this application propose a multi-agent collaborative intelligent response method, a multi-agent collaborative intelligent response device, an electronic device, and a computer-readable storage medium. By unifying and coordinating multiple agents, a multi-agent collaborative architecture capable of dynamically updating the knowledge graph to achieve intelligent response is formed, improving the accuracy and success rate of intelligent response. The core innovation of embodiments of this application lies in constructing a scalable multi-agent architecture that uses natural language to drive the complete lifecycle management of the knowledge graph. This multi-agent architecture consists of multiple modular agents working collaboratively, forming a complete workflow from user natural language input to structured graph response.

[0027] The multi-agent collaborative intelligent response method provided in this application can be applied to terminals and servers, or it can be software running on the server. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The software can be an application that implements the multi-agent collaborative intelligent response method, but it is not limited to the above forms.

[0028] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include server computers, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0029] This application provides a multi-agent collaborative intelligent response method, a multi-agent collaborative intelligent response device, an electronic device, and a computer-readable storage medium. The specific implementation details are provided in the following embodiments. First, the multi-agent collaborative intelligent response method in this application is described.

[0030] It should be noted that in each specific embodiment of this application, when it is necessary to process data related to user identity or characteristics, such as user interface data, the user's permission or consent will be obtained first. Moreover, the collection, use and processing of this data will comply with relevant laws, regulations and standards.

[0031] Reference Figure 1 , Figure 1 This is an optional flowchart of the intelligent response method for multi-agent collaboration provided in the embodiments of this application, which may include, but is not limited to, steps 101 to 107.

[0032] Step 101, Identify the target problem; Step 102: The intent of the target question is obtained by the intent recognition agent. Step 103: Extract concepts from the target problem using a concept extraction agent to obtain target concepts; whereby the target concepts include the first new knowledge. Step 104: The first new knowledge is updated to the preset initial knowledge graph by dynamically updating the intelligent agent to obtain the first knowledge graph; Step 105: The task planning agent performs task planning on the target intent and target concept to obtain a task sequence, which includes the target task. Step 106: Read data from the first knowledge graph through the graph interaction agent and the target task to obtain the target sub-graph data; Step 107: Generate answers to the target sub-map data and target question through the response generating agent to obtain the target answer.

[0033] Steps 101 to 107, as illustrated in this embodiment, operate within a multi-agent collaborative architecture comprising an intent recognition agent, a concept extraction agent, a dynamic update agent, a task planning agent, a graph interaction agent, and a response generation agent. After obtaining the target question, the intent recognition agent outputs the target intent, the concept extraction agent outputs the target concept, the dynamic update agent updates the knowledge graph, the task planning agent outputs the target task, the graph interaction agent outputs the target sub-graph data, and finally, the response generation agent outputs the target answer. Since the knowledge graph is updated by the dynamic update agent before the graph query, there is no need to rely on human expertise to declare new knowledge, and sub-graph data related to the question can be obtained from the knowledge graph. Thus, this application, by unifying and coordinating multiple agents, forms a multi-agent collaborative architecture capable of dynamically updating the knowledge graph to achieve intelligent responses, improving the accuracy and success rate of intelligent responses.

[0034] For example, in a customer service scenario within the fintech field, a customer inputs the question, "Where can I download my transaction statements for the past month?" The intent recognition agent identifies the intent to "download transaction statements." The concept extraction agent extracts concepts from the question, obtaining key concepts such as "transaction," "statement," and "transaction statement." Then, the dynamic update agent updates the knowledge graph, inserting nodes and edges corresponding to "transaction," "statement," and "transaction statement" into the first knowledge graph. Finally, the task planning agent analyzes the intent to "download transaction statements" and the corresponding concepts... The system uses key concepts such as "verifying the target time range (e.g., determining to download the most recent month's bill)," "identifying the account and platform," "verifying customer permissions," and "determining the download operation path." Next, the graph interaction agent queries the knowledge graph of the tasks in the task sequence, obtaining graph query data including "the bill is located on platform A," "the customer has permission to download transaction bills," and "the operation path from the mobile banking homepage to the download page." Finally, the response generation agent answers the question based on the aforementioned graph query data, obtains the target answer, and pushes the target answer to the customer.

[0035] For example, in a patient self-service question-and-answer scenario in the medical field, if a patient enters the question, "What are the early symptoms of high blood sugar?", an intelligent agent can process the question and provide health advice to the patient in easy-to-understand language. The specific working principle of intelligent responses in the medical field is similar to that in the financial field, and will not be elaborated here.

[0036] In step 101 of some embodiments, the target question is obtained. The target question can be spoken or written. When the target question is spoken, it can be acquired through an audio acquisition device such as a microphone. When the target question is written, it can be obtained by converting the spoken question into text, or by obtaining the text input by the user through an input box on the interface. The target question refers to a question raised by the target object, and the target question text is based on natural language, such as "When will the compensation arrive?", "How is my health?", "I want to know what the basic knowledge of deep learning is", "Attention mechanism is a method to improve model performance, how to use attention mechanism to achieve image classification", etc.

[0037] In step 102 of some embodiments, the intent of the target question is identified by an intent recognition agent to obtain the target intent. The intent recognition agent is an agent used to provide intent recognition functionality. The task of the intent recognition agent is to understand the natural language content input by the user and determine which predefined task type the intent belongs to.

[0038] In one example, a user enters an open-ended question through a dialog window, such as "I want to know what the basics of deep learning are." This input can be passed to the large language model in the intent recognition agent for few-shot inference to obtain the target intent. Combining pre-set examples and contextual prompts, the large language model identifies the intent corresponding to the question as "predicting prior knowledge."

[0039] It should be noted that the key to intention recognition intelligent agents is that they do not rely on large-scale supervised data, but instead guide large language models to reason step by step through chain-of-thought prompting technology, enabling them to have strong generalization capabilities and adapt to the complex query needs of various professional fields such as finance, education, healthcare, and law.

[0040] In step 103 of some embodiments, a concept extraction agent extracts concepts from the target question to obtain a target concept. The target concept includes first new knowledge. The concept extraction agent is an agent that provides concept extraction functionality. The concept extraction agent uses natural language processing methods such as Named Entity Recognition (NER) and Relation Extraction (RE) to extract relevant concepts and potential relationships from the user-input question. For example, when a user asks, "Convolutional neural networks are a type of deep learning model, so are convolutional neural networks image processing methods?", the concept extraction agent identifies "convolutional neural network," "deep learning model," and "image processing" as concepts and extracts the semantic relationship "belongs to." The target concept may or may not include first new knowledge. For example, common knowledge graphs often contain the entity "deep learning model" but not necessarily the entity "convolutional neural network." In this case, the concept of "convolutional neural network" can be considered first new knowledge.

[0041] It should be noted that the first and second new knowledge described in this article both refer to entities or semantic relationships that do not exist in the knowledge graph.

[0042] In step 104 of some embodiments, the first new knowledge is updated to a preset initial knowledge graph by a dynamic updating agent, thus obtaining the first knowledge graph. Specifically, if the user-input question involves the addition of new knowledge, the dynamic updating agent can be invoked to update the initial knowledge graph, instead of simply using the initial knowledge graph as a static information source. The dynamic updating agent extracts new entities and relationships from the target question, such as "attention mechanism is a method to improve model performance," identifies the new concept of "attention mechanism," and automatically generates an insertion statement to write it into the initial knowledge graph to obtain the first knowledge graph.

[0043] It should be noted that during the initial knowledge graph update process, deduplication matching, entity renaming, and conflict detection are performed to ensure the integrity and consistency of the graph structure. Dynamically updated intelligent agents allow enterprises or organizations to continuously accumulate and update their private knowledge base while using intelligent response services, thereby constructing a dynamic knowledge graph with evolutionary capabilities.

[0044] In one embodiment, reference is made to Figure 2 Step 104 may include: Step 201: Obtain the number of target concepts to get the number of concepts; Step 202: Determine the new entity based on each target concept, including the first new knowledge; Step 203: If the number of concepts is greater than or equal to 2, then the agent is dynamically updated to extract the first entity relation from the target problem based on any two newly added entities. Step 204: Insert all newly added entities and the first entity relationship into the initial knowledge graph to obtain the first knowledge graph.

[0045] For example, when a user asks, "Convolutional neural networks are a type of deep learning model, so are convolutional neural networks also image processing methods?", the concept extraction agent identifies "convolutional neural network," "deep learning model," and "image processing" as concepts and extracts the semantic relationship "belongs to." "Convolutional neural networks are a type of deep learning model" can be taken as the first new knowledge. Then, both "convolutional neural network" and "deep learning model" are target concepts including this first new knowledge. The agent then dynamically updates its understanding, obtaining entity A based on "convolutional neural network" and entity B based on "deep learning model." It then creates a first semantic relationship between entities A and B based on "belongs to," and finally inserts entities A, B, and the first semantic relationship into the initial knowledge graph to obtain the first knowledge graph.

[0046] The benefits of the embodiments of steps 201 to 204 above are that they realize the natural language construction and dynamic maintenance of professional knowledge graphs, break the dependence of previous graph systems on professional skills, enable non-technical users to participate in the knowledge management process efficiently, and help improve the accuracy and success rate of intelligent responses.

[0047] In one embodiment, reference is made to Figure 3 Step 104 may also include: Step 301: If the number of concepts is equal to 1, the agent is dynamically updated to retrieve entities in the initial knowledge graph based on the target concept, obtain the target associated entity, and determine the second entity relationship between the newly added entity and the target associated entity. Step 302: Insert the newly added entity and the relationship between the second entity into the initial knowledge graph to obtain the first knowledge graph.

[0048] For example, when the user inputs a target question containing "attention mechanism is a method to improve model performance," the concept extraction agent identifies the new concept "attention mechanism." Then, the dynamically updated agent retrieves entities (such as Transformer, image classification, and text generation) in the initial knowledge graph based on "attention mechanism" to obtain the target-related entity, such as Transformer. Next, the dynamically updated agent determines the second semantic relationship between attention mechanism and Transformer to be "containment," meaning Transformer contains attention mechanism. Then, based on the newly added entity "attention mechanism" and the second semantic relationship "containment," an insertion statement is generated to insert the new entity and the second semantic relationship into the initial knowledge graph, resulting in the first knowledge graph.

[0049] The advantage of the embodiments of steps 301 to 302 described above is that if it is not possible to directly obtain two or more entities and the semantic relationships between entities from the target question, new concepts can be inserted into the initial knowledge graph by using entities in the initial knowledge graph, thereby improving the comprehensiveness of the graph update and further helping to improve the accuracy and success rate of intelligent responses.

[0050] In step 105 of some embodiments, a task planning agent performs task planning on the target intent and target concept to obtain a task sequence. The task sequence includes one or more tasks, each of which can be a target task. The goal of the task planning agent is to automatically generate a set of executable target tasks based on the intent and extracted concepts, and to rationally arrange their execution order. For example, if the intent is "pathfinding," the task planning agent will define the first step as locating the starting concept node, the second step as finding the target concept node, and the third step as constructing the shortest path in the graph. The task planning agent uses LLM combined with chained reasoning to decompose and model the tasks, enabling it to automatically adapt to task complexity and support multi-hop queries and conditional dependencies. This planning result serves as an "execution blueprint" for graph operations, providing task structure guidance for subsequent agents.

[0051] In step 106 of some embodiments, the first knowledge graph is read through a graph interaction agent and a target task to obtain target sub-graph data. The graph interaction agent takes the task planning result (i.e., task sequence) as input and automatically constructs a query statement, usually in Cypher language form, based on the current entity nodes and relational structure in the graph, to access or modify the first knowledge graph in the Neo4j graph database.

[0052] In one embodiment, reference is made to Figure 4 Step 106 may include: Step 401: Through the graph interaction agent, generate sentences based on the target concept and entities in the first knowledge graph to obtain the shortest path matching sentences; Step 402: Entity reading is performed on the first knowledge graph based on the shortest path matching statement to obtain the target subgraph data.

[0053] For example, when it is necessary to determine whether two nodes are related, the graph interaction AI can generate a shortest path matching statement and send a request to the graph database to obtain the target subgraph data. For instance, the goal of the shortest path matching statement is to determine whether there is any relationship path between node A and node B with a length not exceeding 3, and return a shortest path pattern.

[0054] The advantage of the embodiments of steps 401 to 402 described above is that they improve the efficiency of map query.

[0055] In one embodiment, step 401 may include: using a BERT-based semantic vector matching mechanism to align the target concept with entities and edges in the first knowledge graph to obtain the first matching node with the closest semantics; and using a graph interaction agent to use the target concept as a second matching node and generate a shortest path matching statement based on the second matching node. This process ensures that subsequent tasks can be performed accurately based on the structured graph data.

[0056] In one embodiment, if the graph interaction agent produces an abnormal result or has no matching item during execution, the intermediate state is fed back to the language model through the ReAct framework, triggering the inference correction process. This human-machine combined feedback mechanism greatly improves the adaptability and robustness of intelligent responses, and is especially suitable for application scenarios with complex graph structures or sparse information.

[0057] It's important to note that ReAct (Reasoning and Acting) is a conversational agent framework that combines reasoning and action. It achieves end-to-end resolution of complex tasks by alternating between reasoning (thinking, planning) and dynamic actions (such as calling tools, querying knowledge bases, and performing tasks) during the conversation. The core idea is to enable the agent to provide not just plain text responses, but also an executable sequence of actions, with each step clearly stating "what I intend to do, why I'm doing it, what information / tools I need, and the expected outcome."

[0058] In one embodiment, prior to step 107, refer to Figure 5 A multi-agent collaborative intelligent response method may further include: Step 501: The reasoning agent performs logical extraction on the target sub-graph data to obtain the target logical chain data; Step 502: The reasoning agent performs task induction based on the target logic chain data and task sequence to obtain task reasoning data. Step 503: Use the task reasoning data as the target sub-graph data.

[0059] Specifically, the role of the reasoning agent is to combine the structured information returned by the graph with the context task to perform semantic integration and logical judgment.

[0060] In one embodiment, reference is made to Figure 6 Step 501 may include: Step 601: Extract the jump paths between entities from the target sub-map data; Step 602: Based on the large language model, information is completed for the concepts and relationships in each jump path to obtain the target optimized path; Step 603: Integrate all target optimization paths to obtain target logical chain data.

[0061] For example, if the graph returns a multi-hop path, the reasoning agent will evaluate whether the path meets the logical requirements of the query target, and at the same time interpret and complete the concepts and relationships in the path to obtain the target optimized path, and then obtain the target logical chain data.

[0062] In step 502, the reasoning agent, after reconstructing the graph structure information into logical chain data, summarizes multi-task and cross-relational reasoning tasks based on the contextual understanding capabilities of the language model. This step is a key indicator of the overall system's intelligence level, ensuring that the intelligent response process not only "finds the right answer" but also "understands it thoroughly," further improving the accuracy of the intelligent response.

[0063] The advantage of the embodiments of steps 501 to 503 described above is that by reorganizing and updating the graph query data, the accuracy of the intelligent response can be improved.

[0064] In one embodiment, after step 502, an intelligent response method based on multi-agent collaboration may further include: extracting concepts from task reasoning data through a concept extraction agent to obtain reasoning concepts; wherein the reasoning concepts include second new knowledge; and updating the first knowledge graph according to the second new knowledge through a dynamic updating agent to obtain a second knowledge graph.

[0065] Specifically, if the dynamic updating agent determines that the reasoning concept includes a second new piece of knowledge, it indicates that new knowledge has been added during task execution. In this case, the dynamic updating agent will be invoked to update the knowledge graph. This feature allows enterprises or organizations to continuously accumulate and update their private knowledge base while using the system, thereby constructing a dynamic knowledge graph with evolutionary capabilities.

[0066] In step 107 of some embodiments, a response generation agent generates an answer to the target subgraph data and the target question, resulting in the target answer. The response generation agent generates a user-readable natural language response. It takes the target subgraph data as input and combines it with the original user input (i.e., the target question) to generate a clearly structured and semantically coherent response statement (i.e., the target answer). For example, for a subgraph matching task, the response generation agent may not only provide "the corresponding node was found," but also output a semantic explanation such as "Tensor decomposition is a type of matrix decomposition, commonly used for dimensionality reduction, and differs from principal component analysis in theoretical methodology." Furthermore, the response generation agent also supports automatic summarization, graph structure description, and task result highlighting, ensuring that users receive intuitive and clear feedback.

[0067] The aforementioned intelligent response method based on multi-agent collaboration can be applied to intelligent response systems. The intelligent response system consists of the aforementioned agents, which interact and collaborate through a unified interface protocol and process control mechanism. Its design emphasizes module decoupling, clear responsibilities, and semantic closure, possessing strong cross-domain adaptability and high interpretability. All agents in the intelligent response system are driven by a Large Language Model (LLM) at their core, but they do not simply call the LLM; instead, the LLM is embedded as a modular component into a well-defined multi-agent system. Each agent focuses on a specific task sub-function, achieving stronger task execution capabilities through reasonable prompting engineering, inference strategies, and process scheduling. This LLM-based multi-agent architecture not only enables human-machine co-construction of knowledge graphs but also provides a new paradigm for task-oriented intelligent response systems.

[0068] Based on the above embodiments, this application can achieve at least the following beneficial effects: (1) Through the multi-agent collaboration mechanism, the intelligent response system has achieved significant improvements in user intent recognition, complex task decomposition, and execution accuracy. (2) The intelligent response system can be seamlessly embedded in privacy-sensitive scenarios such as law and medicine, realizing controllable management of private data, dynamic integration of regulatory updates, and full-process auditing of the reasoning chain. (3) It has significant commercial value and broad industry applicability, and can serve as the core technical support for the next-generation enterprise knowledge management platform.

[0069] Please see Figure 7 This application also provides a multi-agent collaborative intelligent response device, which can implement the above-mentioned multi-agent collaborative intelligent response method. Figure 7The block diagram of the module structure of the intelligent response device for multi-agent collaboration provided in the embodiments of this application includes: a question acquisition module 701, an intent recognition module 702, a concept extraction module 703, a graph update module 704, a task planning module 705, a graph interaction module 706, and a response generation module 707. The system comprises the following modules: a question acquisition module 701, used to acquire a target question; an intent recognition module 702, used to recognize the intent of the target question through an intent recognition agent to obtain the target intent; a concept extraction module 703, used to extract the concept of the target question through a concept extraction agent to obtain the target concept, wherein the target concept includes first new knowledge; a graph update module 704, used to update the first new knowledge to a preset initial knowledge graph through a dynamic update agent to obtain a first knowledge graph; a task planning module 705, used to plan tasks based on the target intent and target concept through a task planning agent to obtain a task sequence, wherein the task sequence includes the target task; a graph interaction module 706, used to read data from the first knowledge graph through a graph interaction agent and the target task to obtain target sub-graph data; and a response generation module 707, used to generate an answer based on the target sub-graph data and the target question through a response generation agent to obtain the target answer.

[0070] In one embodiment, the multi-agent collaborative intelligent response device may further include: a data update module, configured to: extract target sub-graph data logically from the target sub-graph data through the reasoning agent to obtain target logical chain data; perform task induction based on the target logical chain data and task sequence through the reasoning agent to obtain task inference data; and use the task inference data as target sub-graph data.

[0071] In another embodiment, the multi-agent collaborative intelligent response device may further include: a graph optimization module, configured to: extract concepts from task reasoning data through a concept extraction agent to obtain reasoning concepts; wherein the reasoning concepts include second new knowledge; and update the first knowledge graph according to the second new knowledge through a dynamic updating agent to obtain a second knowledge graph.

[0072] It should be noted that the specific implementation of this multi-agent collaborative intelligent response device is basically the same as the specific implementation of the multi-agent collaborative intelligent response method described above, and will not be repeated here.

[0073] This application also provides an electronic device, which includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned intelligent response method for multi-agent collaboration. This electronic device can be any intelligent terminal, including tablet computers, in-vehicle computers, etc.

[0074] Please see Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 to execute the multi-agent collaborative intelligent response method of the embodiments of this application. The 803 input / output interface is used to implement information input and output. The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804); The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0075] This application also provides a computer-readable storage medium for computer-readable storage, wherein the storage medium stores one or more programs, which can be executed by one or more processors to realize the above-described intelligent response method of multi-agent collaboration.

[0076] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0077] The multi-agent collaborative intelligent response method, device, electronic device, and computer-readable storage medium proposed in this application operate on a multi-agent collaborative architecture comprising an intent recognition agent, a concept extraction agent, a dynamic update agent, a task planning agent, a graph interaction agent, and a response generation agent. After obtaining the target question, the intent recognition agent outputs the target intent, the concept extraction agent outputs the target concept, the dynamic update agent updates the knowledge graph, the task planning agent outputs the target task, the graph interaction agent outputs the target sub-graph data, and finally the response generation agent outputs the target answer. Since the knowledge graph is updated by the dynamic update agent before querying the graph, it eliminates the need for manual expertise to declare new knowledge and obtains question-related sub-graph data from the knowledge graph. Thus, this application, by unifying and coordinating multiple agents, forms a multi-agent collaborative architecture capable of dynamically updating the knowledge graph to achieve intelligent response, improving the accuracy and success rate of intelligent responses.

[0078] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0079] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0080] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0082] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0083] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0085] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0087] If the integrated unit is implemented as 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 this application, in essence, 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. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A multi-agent collaborative intelligent response method, characterized in that, The method includes: Identify the target problem; The target intent is obtained by identifying the intent of the target question through an intent-recognition intelligent agent. A concept extraction agent extracts concepts from the target problem to obtain target concepts; wherein the target concepts include first new knowledge. The first knowledge graph is obtained by dynamically updating the first new knowledge to the preset initial knowledge graph through the intelligent agent; A task planning agent performs task planning on the target intent and the target concept to obtain a task sequence, which includes the target task. The target sub-graph data is obtained by reading data from the first knowledge graph through the graph interaction agent and the target task. The response-generating agent generates an answer to the target sub-map data and the target question, thus obtaining the target answer.

2. The method according to claim 1, characterized in that, The step of updating the first new knowledge to a preset initial knowledge graph through a dynamically updating agent to obtain the first knowledge graph includes: Obtain the number of the target concepts to get the number of concepts; The new entity is determined based on each of the target concepts, including the first new knowledge; If the number of concepts is greater than or equal to 2, then the first entity relationship is extracted from the target problem based on any two newly added entities through the dynamically updated agent. All the newly added entities and the relationships between the first entities are inserted into the initial knowledge graph to obtain the first knowledge graph.

3. The method according to claim 2, characterized in that, The step of updating the first new knowledge to a preset initial knowledge graph through a dynamically updating agent to obtain the first knowledge graph further includes: If the number of concepts is equal to 1, the dynamic update agent retrieves entities in the initial knowledge graph based on the target concept to obtain the target associated entity and determines the second entity relationship between the newly added entity and the target associated entity. The newly added entity and the relationship between the second entity are inserted into the initial knowledge graph to obtain the first knowledge graph.

4. The method according to any one of claims 1 to 3, characterized in that, The step of reading data from the first knowledge graph through a graph interaction agent and the target task to obtain target sub-graph data includes: The graph interaction agent generates a statement based on the target concept and the entities in the first knowledge graph to obtain the shortest path matching statement. The first knowledge graph is read based on the shortest path matching statement to obtain the target subgraph data.

5. The method according to any one of claims 1 to 3, characterized in that, Before generating an answer to the target sub-map data and the target question through a response-generating agent to obtain the target answer, the method further includes: The target sub-graph data is logically extracted by the reasoning agent to obtain the target logical chain data; The reasoning agent performs task induction based on the target logical chain data and the task sequence to obtain task reasoning data. The task reasoning data is used as the target sub-map data.

6. The method according to claim 5, characterized in that, The step of logically extracting the target sub-map data through a reasoning agent to obtain target logical chain data includes: Extract the jump paths between entities from the target sub-map data; Based on the large language model, information is filled in for the concepts and relationships in each jump path to obtain the target optimized path; Based on all the target optimization paths, path integration is performed to obtain the target logical chain data.

7. The method according to claim 5, characterized in that, After obtaining task reasoning data by performing task induction through the reasoning agent based on the target logical chain data and the task sequence, the method further includes: The concept extraction agent extracts concepts from the task reasoning data to obtain reasoning concepts; wherein the reasoning concepts include second new knowledge. The dynamically updating agent updates the first knowledge graph based on the second new knowledge to obtain the second knowledge graph.

8. A multi-agent collaborative intelligent response device, characterized in that, The device includes: The problem retrieval module is used to retrieve the target problem; The intent recognition module is used to recognize the intent of the target question through an intent recognition agent to obtain the target intent; A concept extraction module is used to extract concepts from the target problem through a concept extraction agent to obtain target concepts; wherein the target concepts include first new knowledge. The graph update module is used to update the first new knowledge to a preset initial knowledge graph through a dynamic updating agent, thereby obtaining the first knowledge graph; The task planning module is used to perform task planning on the target intent and the target concept through a task planning agent to obtain a task sequence, wherein the task sequence includes the target task. The graph interaction module is used to read data from the first knowledge graph through the graph interaction agent and the target task to obtain target sub-graph data; The response generation module is used to generate an answer to the target sub-map data and the target question through a response generation agent, thereby obtaining the target answer.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images