A multi-source knowledge unified interaction method and system based on a knowledge question answering architecture

Abstract

The application discloses a multi-source knowledge unified interaction method and system based on a knowledge question and answer framework, and the interaction method comprises the following steps: establishing a unified knowledge storage and retrieval system; LLM task disintegration routing; cooperation among multiple intelligent agents through a shared context buffer to form a collaborative knowledge question and answer process; unified authentication; knowledge tracing; constructing a unified knowledge storage system; fusing multi-source data; introducing a multi-agent cooperation mechanism; realizing dynamic permission verification and knowledge tracing; and improving the accuracy and security of question and answer.

Description

Technical Field

[0001] This invention relates to the field of enterprise knowledge management, and in particular to a unified interactive method and system for multi-source knowledge based on a knowledge question-and-answer architecture. Background Technology

[0002] As the application of Large Language Models (LLMs) in enterprise knowledge management becomes increasingly widespread, internal enterprise knowledge is exhibiting a complex coexistence of structured, unstructured, and semi-structured information. Existing enterprise knowledge systems typically rely on a single type of database (such as relational or vector databases) for knowledge management, making it difficult to meet the needs of integrating multimodal and multi-source knowledge.

[0003] In actual business operations, enterprise knowledge types include: Structured data (such as ERP forms, attendance sheets, expense reports, etc.); Unstructured data (such as PDF reports, meeting minutes, emails); Semi-structured data (such as log files, form exports); Multimodal data (such as images, audio recordings, and video-transcribed text).

[0004] Most existing enterprise knowledge question-answering systems adopt the traditional RAG (Retrieval-Augmented Generation) architecture, generating answers through only a single-layer knowledge retrieval and question-answering model. While this approach can improve question-answering performance to some extent, it has significant shortcomings in the following aspects: Unable to handle unified queries from multiple types of knowledge sources; Lack of dynamic access control mechanism; The question-and-answer process is not traceable; Insufficient ability to understand and break down complex tasks at multiple levels; There are "information silos" among different knowledge sources, and there is a lack of a unified semantic fusion mechanism.

[0005] Existing technical solutions: Current mainstream solutions include RAG knowledge question answering systems based on vector databases (such as enterprise knowledge question answering frameworks based on LangChain or LlamaIndex).

[0006] The following process is typically used: Segment and vectorize enterprise documents; Retrieve relevant text using a vector database; Enter the search results into LLM; Output the answer.

[0007] Representative literature, such as "LangChain: Building Applications with LLMs", describes this type of system, which focuses on vector retrieval and context enhancement, but lacks multi-layered collaboration mechanisms and access control mechanisms, making it difficult to apply directly in complex enterprise environments.

[0008] Disadvantages of existing technology: The problem of knowledge silos is severe: structured databases and unstructured knowledge bases operate independently, making unified retrieval and question-and-answer fusion impossible. Access control is coarse-grained: most systems lack dynamic authentication mechanisms, making it impossible to implement differentiated access controls for different roles (employees, supervisors, management). The question-and-answer process is not traceable: existing systems cannot record the source path of questions and answers, making it difficult to support auditing and data governance needs.

[0009] Limited task comprehension: For complex problems, the system struggles to break down tasks and process them in parallel.

[0010] Insufficient multimodal data processing: Existing solutions only support text-based knowledge sources and do not support modal inputs such as voice and images. Summary of the Invention

[0011] In view of the above problems, the present invention is proposed to provide a multi-source knowledge unified interaction method and system based on a knowledge question answering architecture to overcome or at least partially solve the above problems.

[0012] According to one aspect of the present invention, a unified interaction method for multi-source knowledge based on a knowledge question-answering architecture is provided, the interaction method comprising: Establish a unified knowledge storage and retrieval system; LLM task breakdown and routing; Multiple agents collaborate by sharing a context buffer to form a collaborative knowledge question-and-answer process; Unified authentication; Knowledge tracing.

[0013] Optionally, the establishment of a unified knowledge storage and retrieval system specifically includes: Knowledge Fusion Layer: Through self-developed Modular RAG technology, structured databases and unstructured vector databases are merged and stored, adopting a unified knowledge meta-structure; Unified Knowledge Retrieval Engine: Supports semantic-based retrieval and logical condition filtering, and integrates multi-source knowledge results through a multi-path recall mechanism.

[0014] Optionally, the structured database specifically includes: MySQL, time-series database; the multi-path recall mechanism specifically includes: semantic retrieval, keyword matching, and entity association.

[0015] Optionally, the LLM task decomposition routing specifically includes: After receiving the user's question, the task understanding sub-model is invoked to perform semantic analysis and task classification on the input statement; The task is automatically broken down into several sub-tasks based on its type, including calculation, matching, and rule extraction. Distribute subtasks to the corresponding agents in the form of a routing table; After the execution results of each subtask are returned, the results are aggregated and the context is completed.

[0016] Optionally, the collaborative knowledge-based question-answering process formed by the multiple agents cooperating through a shared context buffer specifically includes: Question-answering interaction agent: responsible for engaging in dialogue with users and generating results; Knowledge tracing agent: responsible for verifying the knowledge source of the answer; Permission verification agent: responsible for calling the unified authentication module to verify permissions.

[0017] Optionally, the unified authentication specifically includes: After a user requests access to the system, the unified authentication module is invoked first. The permission template is loaded based on user roles; the user roles include: ordinary employee, department head, and manager. Permission templates are mapped to knowledge source access levels to dynamically adjust search results; A hybrid strategy of role-based access control and attribute-based access control is adopted.

[0018] Optionally, the knowledge tracing specifically includes: Generate a unique TraceID for each question and answer request; The system records the complete chain of request parsing, retrieval path, knowledge retrieval, and answer generation; By establishing a reverse index relationship between source records and knowledge units, a traceable knowledge question-and-answer graph is formed. Based on the report generation logic, a question-and-answer report is automatically generated, including the source document, weight, and confidence level.

[0019] This invention also provides a unified interactive system for multi-source knowledge based on a knowledge question-answering architecture, applying the unified interactive method for multi-source knowledge based on a knowledge question-answering architecture as described in any one of claims 1-7, characterized in that the interactive system comprises: The unified knowledge retrieval engine and knowledge storage system module is used to establish a unified knowledge storage and retrieval system; The LLM task decomposition and routing module is used for LLM task decomposition and routing. The multi-agent collaboration module is used for multiple agents to collaborate through a shared context buffer to form a collaborative knowledge question-and-answer process; The unified authentication module is used for unified authentication. The knowledge tracing module is used for knowledge tracing.

[0020] This invention provides a unified multi-source knowledge interaction method and system based on a knowledge question-answering architecture. The interaction method includes: establishing a unified knowledge storage and retrieval system; LLM task decomposition and routing; collaboration among multiple agents through a shared context buffer to form a collaborative knowledge question-answering process; unified authentication; and knowledge tracing. It constructs a unified knowledge storage system; integrates multi-source data; introduces a multi-agent collaborative mechanism; implements dynamic permission verification and knowledge tracing; and improves the accuracy and security of question answering.

[0021] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a multi-source knowledge unified interaction method based on a knowledge question-answering architecture, provided as an embodiment of the present invention. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0025] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0026] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0027] like Figure 1 As shown, a multi-source knowledge unified interaction method based on a knowledge question-answering architecture includes: Establish a unified knowledge storage and retrieval system; decompose and route LLM tasks; form a collaborative knowledge question-and-answer process through cooperation among multiple agents by sharing a context buffer; unify authentication; and trace knowledge sources.

[0028] This invention proposes a self-developed knowledge base platform and a multi-layered collaborative knowledge question-answering system, consisting of four core modules: (I) Overall Structure The overall system structure includes: Multi-agent collaboration module; LLM task breakdown: routing module; Unified authentication module; Knowledge tracing module; A unified knowledge retrieval engine and knowledge storage system.

[0029] The modules mentioned above collaborate asynchronously through a bus mechanism to decompose, schedule, and integrate the results of knowledge-based question-and-answer tasks.

[0030] (II) Unified knowledge storage and retrieval system 1. Knowledge Integration Layer This layer uses its self-developed Modular RAG technology to integrate and store structured databases (such as MySQL and time-series databases) with unstructured vector databases (VecDB), adopting a unified knowledge meta-structure.

[0031] 2. Unified Knowledge Retrieval Engine This engine supports semantic-based retrieval and logical condition filtering, and integrates multi-source knowledge results through a multi-path recall mechanism (semantic retrieval, keyword matching, entity association).

[0032] (III) LLM Task Decomposition Routing Module 1. After receiving a user's question, the module calls the task understanding sub-model to perform semantic analysis and task classification on the input statement.

[0033] 2. Automatically break down tasks into several sub-tasks based on task type (such as calculation, matching, rule extraction).

[0034] 3. Distribute the subtasks to the corresponding agents in the form of a routing table.

[0035] 4. After the execution result of each subtask is returned, the routing module is responsible for result aggregation and context completion.

[0036] (iv) Multi-agent collaboration module The system is configured with three types of core intelligent agents: Question-answering interactive intelligent agent: responsible for engaging in dialogue with users and generating results; Knowledge tracing agent: responsible for verifying the knowledge source of the answer; Permission verification agent: responsible for calling the unified authentication module to verify permissions.

[0037] Multiple agents collaborate by sharing a context buffer, forming a "collaborative knowledge question answering" process.

[0038] (v) Unified Authentication Module 1. After a user requests access to the system, the unified authentication module is invoked first; 2. The module loads permission templates based on user roles (regular employee / department leader / manager); 3. Map permission templates to knowledge source access levels to dynamically adjust search results; 4. This module adopts a hybrid strategy of role-based access control (RBAC) and attribute-based access control (ABAC).

[0039] (vi) Knowledge Tracing Module 1. Generate a unique TraceID for each question-and-answer request; 2. The system records the complete chain of request parsing, retrieval path, knowledge retrieval, and answer generation; 3. Establish a reverse index relationship between the source records and knowledge units to form a traceable knowledge question-and-answer graph; 4. Based on the report generation logic, automatically output a Q&A report, including source documents, weights, and confidence levels.

[0040] (vii) Overall steps of the Q&A process 1. The user initiates a question-and-answer request; 2. System authentication; 3. Task breakdown; 4. Collaborative execution by intelligent agents; 5. Call the knowledge retrieval engine; 6. Aggregation results; 7. Output a source-tracing answer and question-and-answer report.

[0041] Beneficial effects: This paper provides a knowledge base platform integrating structured and unstructured data and its multi-layered collaborative knowledge question answering method, which improves the accuracy and security of question answering by: constructing a unified knowledge storage system; integrating multi-source data; introducing a multi-agent collaborative mechanism; realizing dynamic permission verification and knowledge traceability; and improving the accuracy and security of question answering.

[0042] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A unified interactive method for multi-source knowledge based on a knowledge question-answering architecture, characterized in that, The interaction method includes: Establish a unified knowledge storage and retrieval system; LLM task breakdown and routing; Multiple agents collaborate by sharing a context buffer to form a collaborative knowledge question-and-answer process; Unified authentication; Knowledge tracing.

2. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 1, characterized in that, The establishment of a unified knowledge storage and retrieval system specifically includes: Knowledge Fusion Layer: Through self-developed Modular RAG technology, structured databases and unstructured vector databases are merged and stored, adopting a unified knowledge meta-structure; Unified Knowledge Retrieval Engine: Supports semantic-based retrieval and logical condition filtering, and integrates multi-source knowledge results through a multi-path recall mechanism.

3. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 2, characterized in that, The structured database specifically includes: MySQL and time-series databases; the multi-path recall mechanism specifically includes: semantic retrieval, keyword matching, and entity association.

4. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 1, characterized in that, The LLM task decomposition routing specifically includes: After receiving the user's question, the task understanding sub-model is invoked to perform semantic analysis and task classification on the input statement; The task is automatically broken down into several sub-tasks based on its type, including calculation, matching, and rule extraction. Distribute subtasks to the corresponding agents in the form of a routing table; After the execution results of each subtask are returned, the results are aggregated and the context is completed.

5. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 1, characterized in that, The collaborative knowledge-based question-answering process formed by the multi-agent collaboration through a shared context buffer specifically includes: Question-answering interaction agent: responsible for engaging in dialogue with users and generating results; Knowledge tracing agent: responsible for verifying the knowledge source of the answer; Permission verification agent: responsible for calling the unified authentication module to verify permissions.

6. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 1, characterized in that, The unified authentication specifically includes: After a user requests access to the system, the unified authentication module is invoked first. The permission template is loaded based on user roles; the user roles include: ordinary employee, department head, and manager. Permission templates are mapped to knowledge source access levels to dynamically adjust search results; A hybrid strategy of role-based access control and attribute-based access control is adopted.

7. The multi-source knowledge unified interaction method based on a knowledge question-answering architecture according to claim 1, characterized in that, The knowledge tracing specifically includes: Generate a unique TraceID for each question and answer request; The system records the complete chain of request parsing, retrieval path, knowledge retrieval, and answer generation; By establishing a reverse index relationship between source records and knowledge units, a traceable knowledge question-and-answer graph is formed. Based on the report generation logic, a question-and-answer report is automatically generated, including the source document, weight, and confidence level.

8. A multi-source knowledge unified interaction system based on a knowledge question-answering architecture, employing the multi-source knowledge unified interaction method based on a knowledge question-answering architecture as described in any one of claims 1-7, characterized in that, The interactive system includes: The unified knowledge retrieval engine and knowledge storage system module is used to establish a unified knowledge storage and retrieval system; The LLM task decomposition and routing module is used for LLM task decomposition and routing. The multi-agent collaboration module is used for multiple agents to collaborate through a shared context buffer to form a collaborative knowledge question-and-answer process; The unified authentication module is used for unified authentication. The knowledge tracing module is used for knowledge tracing.

Images