Intelligent knowledge management method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computers, in particular to an intelligent knowledge management method and system, electronic equipment and a storage medium. The method comprises the steps that a unified knowledge base is constructed, structured data, semi-structured files and non-structured files are imported from a heterogeneous data source, a multi-dimensional classification system with a father-son relation and a level relation is established, and classified knowledge data is generated; processing the knowledge data, including layout analysis, structured processing and semantic information extraction, to generate processed knowledge data; converting the processed knowledge data into a vector form, constructing a vector knowledge base, based on the vector knowledge base, combining AI model capability and a plug-in tool through dynamic workflow, providing a knowledge application function and a domain agent service, capturing user derived knowledge and returning the knowledge to the vector knowledge base, and forming a knowledge closed loop. According to the method, the agility, intelligence and expandability of knowledge management can be met.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and specifically to an intelligent knowledge management method, system, electronic device, and storage medium. Background Technology

[0002] Currently, knowledge management systems are primarily built and optimized around basic functional modules, forming a basic functional framework that includes knowledge base construction and management, intelligent retrieval, and question-answering services. Although the system's capabilities have been iteratively upgraded in recent years through the introduction of technologies such as large language models, significant technical bottlenecks still exist in practical applications. Specifically:

[0003] There is a serious disconnect between the data processing layer and the upper intelligent application layer, resulting in low efficiency of knowledge flow; the capabilities of artificial intelligence models and intelligent agents are relatively fixed, making it difficult to flexibly adjust according to complex and ever-changing business needs, forming the phenomenon of "intelligent islands"; the coupling between various functional modules is too high, the system's scalability is limited, and it is difficult to quickly migrate and customize to different vertical fields.

[0004] Especially when addressing users' personalized knowledge management needs, existing systems often require significant manpower for customization, and the modified functional modules are difficult to reuse effectively in other application scenarios. This situation severely restricts the rapid deployment and application effectiveness of knowledge management systems in industry-level solutions, failing to meet the agility, intelligence, and scalability requirements of modern enterprises for knowledge management.

[0005] Therefore, this application provides an intelligent knowledge management method to solve the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent knowledge management method, system, electronic device, and storage medium to solve the technical problems in the prior art that cannot meet the requirements for agility, intelligence, and scalability in knowledge management.

[0007] To address the aforementioned technical problems, this invention provides an intelligent knowledge management method, comprising:

[0008] Build a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data;

[0009] The knowledge data is processed, including layout parsing, structure processing, and semantic information extraction, to generate processed knowledge data;

[0010] The processed knowledge data is transformed into vector form to build a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation.

[0011] Based on the aforementioned vector knowledge base, AI model capabilities and plug-in tools are combined through a dynamic workflow to provide knowledge application functions and domain intelligent agent services, and user-derived knowledge is captured and fed back to the vector knowledge base to form a knowledge closed loop.

[0012] In some specific embodiments, a unified knowledge base is constructed, importing structured data, semi-structured files, and unstructured documents from heterogeneous data sources, establishing a multi-dimensional classification system with parent-child and peer relationships, and generating classified knowledge data, further including:

[0013] Import structured data from a relational database, establish a connection with the source database, and extract the data table structure and records;

[0014] Import data from semi-structured files, parse the file format, and extract structured information;

[0015] Import data from unstructured documents, identify document types, and extract text content;

[0016] Establish a multi-dimensional classification system with parent-child and peer-level relationships. The classification system includes classification perspectives based on technical field, product line, project affiliation, business process stage, knowledge type, applicable role, and geographical region.

[0017] The automatic classification task connects knowledge data with the classification system, generating classified knowledge data.

[0018] In some specific embodiments, the knowledge data is processed, including layout parsing, structured processing, and semantic information extraction, to generate processed knowledge data, further including:

[0019] Perform layout analysis to identify the physical layout and visual structure of the document, and locate the content areas and spatial relationships within the document;

[0020] The text blocks obtained from page layout parsing are converted into structured data with clearly defined fields, formats, and heading hierarchies.

[0021] Information extraction is performed to extract deep semantic information from text, including entity recognition and relation extraction.

[0022] The system stores the processing results, including the results of layout parsing, structured processing, and information extraction.

[0023] In some specific embodiments, the processed knowledge data is transformed into vector form to construct a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation, and further includes:

[0024] Perform format conversion to unify the heterogeneous raw data in the knowledge base into a standard text format suitable for data slicing;

[0025] Data slicing is performed, and standard text is divided into semantically complete text fragments based on the processing capabilities of the vectorization model and the needs of downstream applications.

[0026] Vector generation is performed, and a vectorization model is used to convert text fragments into vector representations to form a vector knowledge base;

[0027] Vector validation is performed, and the quality of the generated vectors is evaluated through similarity validation and question-answering source tracing validation.

[0028] In some specific embodiments, providing knowledge application functions and domain intelligent agent services further includes:

[0029] It provides intelligent question-answering functionality, which identifies the user's question intent and automatically retrieves other functions;

[0030] It provides intelligent search functionality to retrieve information for question-and-answer sessions and to provide materials for writing and reading.

[0031] It provides intelligent writing and intelligent reading functions, and calls retrieval and question-and-answer functions during the execution process, and stores the generated content in the knowledge base;

[0032] It provides knowledge graph and personal knowledge management functions, provides underlying data support, and collects the generated knowledge;

[0033] Construct domain-specific intelligent agent services, including domain-specific question-answering, report generation, chart analysis, document review, and information extraction intelligent agents.

[0034] In some specific embodiments, building a domain intelligence agent service includes:

[0035] Construct a domain-specific question-answering intelligent agent based on a retrieval-enhanced generative architecture, combined with semantic retrieval and a generative large model;

[0036] Construct a report generation intelligent agent that integrates text summarization, table generation, and data visualization components based on multimodal data fusion and template generation;

[0037] Construct a chart analysis-type intelligent agent based on a data-driven analysis engine, integrating statistical modeling, time series forecasting, and computer vision technologies;

[0038] Construct a document review-type intelligent agent, combining a rule engine and a deep learning model, to achieve text compliance checks, risk identification, and version comparison;

[0039] Construct an information extraction intelligent agent to support users in defining extraction fields, relationships, and output formats, thereby transforming unstructured and semi-structured data into structured results.

[0040] In some specific embodiments, the method further includes the step of building a vertical domain knowledge management application:

[0041] Build a vertical domain vector library, and achieve automated data processing by constructing a knowledge library, knowledge system, knowledge processing and extraction, and knowledge vectorization in a layered manner;

[0042] It integrates intelligent orchestration of multiple models, and manages model capabilities in a unified manner through a model repository and API gateway. All model capabilities are exposed to the outside world through a unified and standardized API gateway.

[0043] Integrate knowledge application functions, using knowledge graphs as a hub to create a synergistic effect;

[0044] Build vertical domain intelligent agents and perform in-depth optimization and configuration for specific domain tasks.

[0045] Based on the same concept, the present invention also provides an intelligent knowledge management system, comprising:

[0046] The knowledge base construction module is configured to build a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data.

[0047] The data processing module is configured to process the knowledge data, including layout parsing, structure processing and semantic information extraction, to generate processed knowledge data;

[0048] The vector transformation module is configured to convert processed knowledge data into vector form and build a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation.

[0049] The knowledge management and optimization module is configured to provide knowledge application functions and domain intelligent agent services based on the vector knowledge base, through dynamic workflow combination of AI model capabilities and plug-in tools, and capture user-derived knowledge to flow back to the vector knowledge base, forming a knowledge closed loop.

[0050] Based on the same concept, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of an intelligent knowledge management method.

[0051] Based on the same concept, the present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of an intelligent knowledge management method.

[0052] Compared with existing technologies, its advantages are as follows:

[0053] This invention discloses an intelligent knowledge management method, system, electronic device, and storage medium. Through a four-layer subsystem architecture design, it achieves hierarchical separation of data governance, AI capability provision, application services, and scenario implementation. This solves the problems of fragmented functional modules and redundant construction in traditional systems, improving the system's flexibility, maintainability, and scalability.

[0054] Intelligent Capability Agile Delivery: Based on a multi-model intelligent orchestration subsystem, a dynamic workflow mechanism is adopted to achieve flexible combination and scheduling of AI capabilities. Breaking the limitations of "intelligent silos," it can quickly respond to complex and ever-changing business needs and efficiently execute complex tasks.

[0055] A self-evolving knowledge loop system: It can not only process raw knowledge data, but also capture and structure the derivative knowledge generated by users in the application process in real time, and realize the automatic backflow and updating of knowledge through the data foundation layer, forming a knowledge loop system that continuously learns and improves itself.

[0056] Vertical domain migration capability: The generalized underlying subsystem architecture provides infrastructure support for rapid migration to different vertical domains. Through infrastructure reuse and domain capability assembly, it achieves the effect of "build once, reuse in multiple domains", reducing the cost and time of customized development.

[0057] With intelligent question answering as the core entry point, it achieves organic synergy between functions such as intelligent retrieval, intelligent writing, intelligent reading, knowledge graphs, and personal knowledge management, forming a unified knowledge application ecosystem and improving user experience and system efficiency. Attached Figure Description

[0058] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0059] Figure 1 This is a flowchart illustrating some specific embodiments of the intelligent knowledge management method of the present invention;

[0060] Figure 2 This is one of the flowcharts of another embodiment of the intelligent knowledge management method of the present invention;

[0061] Figure 3 This is a second flowchart illustrating another embodiment of the intelligent knowledge management method of the present invention;

[0062] Figure 4 This is a third flowchart illustrating another embodiment of the intelligent knowledge management method of the present invention;

[0063] Figure 5 This is the fourth flowchart of another embodiment of the intelligent knowledge management method of the present invention;

[0064] Figure 6 This is the fifth flowchart of another embodiment of the intelligent knowledge management method of the present invention;

[0065] Figure 7 This is a schematic diagram of the structure of an intelligent knowledge management system according to some specific embodiments of the present invention;

[0066] Figure 8 This is a schematic diagram of the structure of an electronic device according to some specific embodiments of the present invention;

[0067] In the diagram, 710 is the processor; 720 is the memory; 730 is the input device; and 740 is the output device. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0070] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0071] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0072] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0073] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0074] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0075] Reference Figure 1 An intelligent knowledge management method includes:

[0076] S101, Construct a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data;

[0077] S102, The knowledge data is processed, including layout parsing, structure processing and semantic information extraction, to generate processed knowledge data;

[0078] S103, transform the processed knowledge data into vector form and construct a vector knowledge base. The transformation process includes format conversion, data slicing and vector generation.

[0079] S104. Based on the vector knowledge base, AI model capabilities and plug-in tools are combined through a dynamic workflow to provide knowledge application functions and domain intelligent agent services, and user-derived knowledge is captured and fed back to the vector knowledge base to form a knowledge closed loop.

[0080] Specifically, in this embodiment of the invention, structured data, semi-structured files, and unstructured documents are imported from heterogeneous data sources. Structured data is imported by establishing data connections and extracting table structures; semi-structured files are parsed to extract structured information; and unstructured documents are identified by type recognition to extract text content. A multi-dimensional classification system with parent-child and peer relationships is established. This system includes multiple classification dimensions, each representing a different knowledge organization perspective. Knowledge data is linked to the classification system through an automatic classification task, generating knowledge data with multi-dimensional classification identifiers. The classified knowledge data is then processed, including using computer vision and optical character recognition technologies for layout analysis to identify document layout structure and spatial relationships, and converting the parsed text into text with clearly defined fields and... The hierarchical data is processed by extracting entities and their relationships from the text through semantic information extraction. The processed knowledge data is then transformed into vector form. This process includes format conversion, which unifies heterogeneous data into a standard text format; data slicing, which divides the text into semantically complete segments according to processing requirements; and vector generation, which uses a vectorization model to transform text segments into high-dimensional numerical vectors, thus constructing a vector knowledge base. Based on this vector knowledge base, various AI model capabilities and plug-in tools are combined through a dynamic workflow mechanism to achieve task decomposition, data flow, and result aggregation. This provides knowledge application functions such as intelligent question answering, intelligent retrieval, and intelligent writing, as well as intelligent agent services in multiple domains. During the service process, derivative knowledge generated by user interaction is captured, structured, and then fed back into the vector knowledge base, forming a continuously evolving knowledge closed-loop system.

[0081] For example, in a contract intelligent review scenario, when building a unified knowledge base, structured data from basic contract information tables is imported from an SQL Server database, semi-structured data from contract template clauses is imported from Excel files, and unstructured data from historical contract texts is imported from PDF documents. The established multi-dimensional classification system includes four dimensions: contract type, contracting parties, contract amount, and performance period. The contract type dimension includes subcategories such as purchase contracts, sales contracts, and service contracts, while the contracting parties dimension is divided geographically into North China, East China, and South China. Contract data is automatically linked according to these dimensions through an automatic classification task, generating a contract knowledge base with multi-dimensional classification identifiers. In the processing stage, PDF contracts are parsed to identify the positional relationships of elements such as clause titles, body text, and signature areas. Through structured processing, contract clauses are converted into structured data containing fields such as clause type, effective conditions, and liability for breach of contract. Semantic information is used to extract entities and their relationships, such as the contracting parties, contract amount, and performance period. In the next step... During the quantification process, contract texts are uniformly converted to UTF-8 encoded plain text format. Data is sliced ​​at a granularity of 500 characters per segment, and the BERT model is used to transform text fragments into 768-dimensional numerical vectors, constructing a contract vector knowledge base. Based on this vector base, a dynamic workflow is used to combine text understanding models, rule engines, and auditing plugins to provide services such as intelligent contract retrieval, clause comparison, and risk warnings. When users mark new risk clauses during use, this information is captured, and risk features are extracted through semantic analysis. New vector representations are generated and updated to the vector knowledge base, achieving continuous improvement of the knowledge system. Specifically, in similarity calculation, the cosine similarity algorithm is used to calculate the cosine value of the angle between the vector of a new clause and an existing clause in the knowledge base. When the similarity reaches a set threshold, it is automatically classified into the corresponding risk category. For example, if the calculated similarity is 0.85, exceeding the threshold of 0.8, the clause is classified into the payment default risk category.

[0082] In some applications, a unified knowledge base is constructed, importing structured data, semi-structured files, and unstructured documents from heterogeneous data sources. A multi-dimensional classification system with parent-child and peer-level relationships is established to generate classified knowledge data. This includes importing structured data from relational databases, establishing connections with the source databases, and extracting data table structures and records; importing data from semi-structured files, parsing file formats, and extracting structured information; importing data from unstructured documents, identifying document types, and extracting text content; establishing a multi-dimensional classification system with parent-child and peer-level relationships, which includes classification perspectives such as technical field, product line, project affiliation, business process stage, knowledge type, applicable role, and geographical region; and connecting knowledge data with the classification system through automatic classification tasks to generate classified knowledge data.

[0083] Understandably, importing structured data from a relational database involves establishing a database connection to access the source database system, obtaining metadata information such as table structure definitions, field types, and constraints, and extracting complete database record content. When importing data from semi-structured files, the process involves identifying the file's storage format characteristics, parsing the internal data organization structure, identifying key data fields and their hierarchical relationships, and extracting the structured information contained within. When importing data from unstructured documents, a document type recognition engine determines the document format category, and content extraction techniques are used to obtain the text information within the document, preserving the semantic integrity of the original document. A multi-dimensional classification system with parent-child and peer relationships is established, encompassing technical dimensions including basic theories, applied technologies, experimental methods, and other classification perspectives. The system employs various classification dimensions, including: linear dimensions (core products, auxiliary products, service products, etc.), project affiliation dimensions (projects under development, completed projects, projects awaiting approval, etc.), business process stage dimensions (requirements analysis, solution design, development implementation, testing and verification, etc.), knowledge type dimensions (technical documents, research reports, standards and specifications, experience summaries, etc.), applicable role dimensions (R&D personnel, testers, managers, customer representatives, etc.), and geographical region dimensions (domestic and international regions, etc.). Through automatic classification tasks, imported knowledge data is matched with the multi-dimensional classification system. Similarity calculations are performed based on the feature attributes of the knowledge data and the definition rules of the classification system, enabling automatic linking of knowledge data across multiple classification dimensions and generating knowledge data with complete classification identifiers.

[0084] For example, in building a knowledge base for the field of intelligent vehicle technology, during the stage of importing structured data, the product data table in the MySQL database was connected to extract 1250 product records containing fields such as product number, product name, technical parameters, and development status; during the stage of importing semi-structured data, JSON-formatted technical document files were parsed to extract structured fields such as document title, author information, key technical points, and experimental data, processing a total of 280 technical documents; during the stage of importing unstructured data, PDF-formatted research reports and Word-formatted design documents were identified, extracting complete text content, including technical solution descriptions, experimental result analysis, conclusions, and recommendations, processing a total of 150 research reports and 95 design documents; in the established multi-dimensional classification system, the technology field dimension... The system is divided into three primary categories: intelligent driving, connected vehicles, and new energy vehicles. Intelligent driving is further subdivided into three secondary categories: environmental perception, decision-making and planning, and control execution. At the product line level, it includes five parallel categories: in-vehicle systems, intelligent terminals, cloud service platforms, etc. At the project affiliation level, it has three hierarchical categories based on project status. Through automatic classification tasks, the system calculates the similarity between keyword features of knowledge content and category tags. For example, a document titled "Application of Millimeter-Wave Radar in Autonomous Driving" has a similarity of 0.92 to the intelligent driving category at the technology level and 0.85 to the in-vehicle systems category at the product line level. These similarities are automatically linked to the corresponding category nodes, ultimately generating 1850 knowledge data records with complete category identifiers, forming a classification knowledge base for subsequent processing.

[0085] In some applications, the knowledge data is processed, including layout parsing, structuring, and semantic information extraction, to generate processed knowledge data. This includes performing layout parsing to identify the physical layout and visual structure of the document, and locating content areas and spatial relationships within the document; performing structuring to convert the text blocks obtained from layout parsing into structured data with clearly defined fields, formats, and heading hierarchies; performing information extraction to extract deep semantic information from the text, including entity recognition and relation extraction; and storing the processing results, which involve uniformly storing the results of layout parsing, structuring, and information extraction.

[0086] Understandably, the process involves: 1) Page layout analysis: using computer vision technology to analyze the document's page layout features, identifying the boundary positions and spatial distribution relationships of various content areas, and establishing a coordinate positioning system for document elements; 2) Structured processing: transforming the text blocks obtained from page layout analysis according to a predefined data model to generate structured data with clearly defined fields, format specifications, and heading hierarchy relationships, thus establishing a complete content organization structure; 3) Information extraction: using natural language processing technology to identify and extract entity units with specific semantics from the text content, while analyzing the relationships between entities to construct a semantic association network; and 4) Integrating and storing the layout information obtained from page layout analysis, the data structure generated from structured processing, and the semantic content extracted from information to establish a complete database of processed results.

[0087] For example, during page layout analysis, a PDF document was identified as containing five types of content areas: title area, abstract block, body paragraphs, data tables, and chart descriptions. The title area is located at the top of the page within a Y-coordinate range of 120-150 pixels, the abstract block within a Y-coordinate range of 160-220 pixels, and the body paragraphs within a Y-coordinate range of 230-650 pixels. During structured processing, the identified text blocks are converted into structured data containing fields such as document title, author information, abstract content, technical parameters, and implementation cases. The title hierarchy is set to one first-level heading and two second-level headings. There are 3 questions and 8 third-level headings. During information extraction, 35 technical entities such as "neural network", "deep learning" and "convolution calculation" were identified from the main text, and 12 sets of entity relationships such as "neural network applied to image recognition" and "deep learning optimization algorithm" were extracted. When storing the processing results, the coordinate information of the page layout, the field data of the structured processing, and the entity relationship table of information extraction were uniformly stored in the database. The entity recognition accuracy reached 92.5%, the relationship extraction completeness reached 88.7%, and the resulting processing result data was stored as a total of 125 structured records.

[0088] In some applications, processed knowledge data is transformed into vector form to build a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation. Format conversion involves unifying the heterogeneous raw data in the knowledge base into a standard text format suitable for data slicing. Data slicing involves dividing the standard text into semantically complete text fragments based on the processing capabilities of the vectorization model and the needs of downstream applications. Vector generation involves using a vectorization model to convert the text fragments into vector representations, forming a vector knowledge base. Vector verification involves evaluating the quality of the generated vectors through similarity verification and question-answering source tracing verification.

[0089] Understandably, the process involves: format conversion, transforming raw data in different formats within the knowledge base into a standardized text format using unified conversion rules to ensure data consistency; data slicing, dividing the standardized text into appropriately sized segments based on semantic boundaries according to the maximum processing length limit of the selected vectorization model and the semantic integrity requirements of downstream applications; vector generation, using a pre-trained vectorization model to map text segments into numerical vectors in a high-dimensional space, which capture the deep semantic features of the text; and finally, vector quality verification, comparing the vector distances of related text segments through similarity verification, checking the actual performance of vectors in question-answering tasks through question-answering source tracing verification, and optimizing and improving the vector knowledge base based on the verification results.

[0090] For example, in the format conversion stage, 85 technical reports containing a mix of tables, charts, and text were uniformly converted into plain text format, preserving the original paragraph structure and data relationships. In the data slicing stage, based on the BERT model's processing limit of 512 tokens, each report was divided into multiple text segments by chapter, with each segment containing 3-5 semantically coherent paragraphs, generating a total of 1250 text segments. In the vector generation stage, the BERT-base model was used to convert each text segment into a 768-dimensional floating-point vector, with each dimension representing a different semantic feature, ultimately forming a knowledge base containing 1250 vectors. In the vector validation stage, 10 groups of semantically similar technical report segments were selected for similarity validation, and the cosine similarity of each group of segment vectors was calculated. The results showed that the similarity was all above 0.8, meeting the expected standard. Through question-and-answer source tracing validation, 50 technical questions were used to test the retrieval effect of the vector library. The results showed that the vector-based retrieval accuracy reached 89.3%, and the recall rate reached 85.7%, validating the reliability of the vector quality.

[0091] Some of these applications provide knowledge application functions and domain-specific intelligent agent services, including intelligent question-answering functions that identify user question intent and automatically retrieve other functions; intelligent retrieval functions that provide information retrieval for question-answering and supply materials for writing and reading; intelligent writing and reading functions that invoke retrieval and question-answering during execution, and store the generated content in a knowledge base; knowledge graph and personal knowledge management functions that provide underlying data support and collect the generated knowledge; and the construction of domain-specific intelligent agent services, including domain-specific question-answering, report generation, chart analysis, document review, and information extraction intelligent agents.

[0092] Understandably, the system offers intelligent question-answering functionality, using natural language understanding technology to analyze user query intent and automatically calling relevant functional modules to form a processing chain based on intent recognition results; intelligent retrieval functionality, employing a multi-path recall mechanism to provide relevant information support for the question-answering process, while also providing data supply services for writing and reading functions; intelligent writing and reading functionality, using real-time retrieval functions to obtain reference materials during document writing, embedding question-answering functionality during document reading to achieve interactive understanding, and automatically storing newly generated content in the knowledge base; knowledge graph and personal knowledge management functionality, constructing a semantic association network to provide underlying data support for upper-layer applications, and establishing a personal knowledge space to store the knowledge output of various functional modules; and building domain-specific intelligent agent services, developing domain-specific question-answering intelligent agents based on a modular architecture to handle professional domain-specific inquiries, report generation intelligent agents to automatically synthesize structured documents, chart analysis intelligent agents to perform data visualization analysis, document review intelligent agents to perform content compliance checks, and information extraction intelligent agents to complete the structured transformation of unstructured data.

[0093] For example, in the field of scientific research management, in the implementation of the intelligent question-answering function, when a user enters the query "the latest applications of deep learning in medical imaging," the system identifies this as a technical research intent, automatically invokes the search function to obtain relevant literature, and calls the report generation function to organize the data, returning a complete answer containing 15 references within 0.8 seconds. In the implementation of the intelligent search function, a three-pronged approach of keyword search, vector search, and graph search is used to retrieve 35 relevant documents for the question-answering stage and 28 technical cases for intelligent writing, achieving a search accuracy rate of 89.5%. In the implementation of the intelligent writing and reading functions, when a user writes an "AI medical review," the system automatically invokes the search function to obtain 32 of the latest materials, and when reading the literature... The question-and-answer function analyzed professional terms 46 times, and the generated 3 research reports were automatically stored in the knowledge base. In the implementation of knowledge graph and personal knowledge management functions, a medical imaging knowledge graph containing 1,250 entities and 2,800 relationships was constructed, and 85 documents and 240 notes generated by users were stored in the personal knowledge space. In the implementation of domain intelligent agent services, the domain question-and-answer intelligent agent processed 156 professional consultations, the report generation intelligent agent automatically generated 28 technical reports, the chart analysis intelligent agent analyzed 65 medical images, the document review intelligent agent completed compliance checks on 89 documents, and the information extraction intelligent agent extracted 1,250 pieces of structured data from 230 documents, forming a complete knowledge service closed loop.

[0094] In some of these applications, the construction of domain-specific intelligent agents includes: building domain-specific question-answering agents based on a retrieval-enhanced generative architecture, combined with semantic retrieval and generative large models; building report generation agents based on multimodal data fusion and template-based generation, integrating text summarization, table generation, and data visualization components; building chart analysis agents based on a data-driven analysis engine, integrating statistical modeling, time series prediction, and computer vision technologies; building document review agents, combining rule engines and deep learning models to achieve text compliance checks, risk identification, and version comparison; and building information extraction agents to support user-defined extraction fields, relationships, and output formats, realizing the transformation of unstructured and semi-structured data into structured results.

[0095] Understandably, the construction of a domain-specific question-answering agent employs a retrieval-enhanced generative architecture, collaboratively integrating a semantic retrieval module with a generative large-scale model. Semantic retrieval extracts relevant information fragments from the knowledge base, and the generative large-scale model further processes the retrieval results and generates natural language responses. A report generation agent utilizes a multimodal data fusion mechanism to uniformly represent data from different sources and formats, combines template-based generation technology to define the report's structural framework, integrates a text summarization component for content condensation, a table generation component for data organization, and a data visualization component for graphical display. A chart analysis agent is constructed based on a data-driven analysis engine to extract features and recognize patterns from input data, integrates statistical modeling methods for data distribution analysis and correlation verification, and uses time-series... Predictive technologies are used for trend analysis and future value prediction, while computer vision technologies are integrated for image feature extraction and pattern recognition. A document review agent is constructed, organically combining a rule engine based on predefined rules with a deep learning-based neural network model. The rule engine enables structured checks of explicit knowledge, while the deep learning model identifies and mines implicit patterns, jointly completing text compliance verification, risk point identification, and comparison of differences between different versions. An information extraction agent is also constructed, providing a user-configurable field definition interface that allows users to independently set the types of information fields to be extracted according to specific needs, define semantic relationships between entities, and specify the data form and structure format of the output results, achieving automated conversion from unstructured raw data and semi-structured source data to regularized structured data.

[0096] For example, in the field of financial risk control, when building a domain-specific question-answering agent, for the query "identify transaction fraud characteristics," semantic retrieval retrieved 15 relevant legal provisions and 28 historical cases from the knowledge base. A generative large-scale model was then used to synthesize a complete answer containing 5 major categories of features and 12 specific indicators, achieving a question-answering accuracy of 95.3%. In building a report generation agent, it processed multimodal risk control data including 85 tables, 23,000 words of text description, and 24 trend charts. Following a standard risk control report template, it automatically generated a structured report containing text summaries, statistical tables, and visualization charts, achieving a report generation completeness of 92.7%. In building a chart analysis agent, statistical analysis of 150 transaction datasets revealed an abnormal transaction rate of 7.8%. The predictive model achieved an accuracy of 88.5% in predicting risk trends over the next 30 days and identified 12 types of suspicious transaction patterns using computer vision technology. In building a document review agent, the rule engine set 35 compliance check rules, and the deep learning model was trained using 100,000 labeled data points. It reviewed 500 loan contracts, identifying 28 compliance issues, labeling 15 risk clauses, and discovering 9 significant changes through version comparison. In building an information extraction agent, users customized extraction fields including 8 core fields such as customer identity information, transaction amount, and timestamp, defined 3 types of relationships, and specified JSON format as the output. It extracted structured data records from 2,856 unstructured transaction records with an accuracy of 94.2% and a completeness rate of 91.8%.

[0097] In some applications, the method further includes the steps of building vertical domain knowledge management applications: building a vertical domain vector library, achieving automated data processing through layered knowledge library construction, knowledge system, knowledge processing and extraction, and knowledge vectorization; integrating multi-model intelligent orchestration, managing model capabilities uniformly through model repositories and API gateways, with all model capabilities exposed externally through a unified standardized API gateway; integrating knowledge application functions, forming a linkage effect with knowledge graphs as the hub; and building vertical domain intelligent agents for in-depth optimization and configuration for specific domain tasks.

[0098] Understandably, the construction of a vertical domain vector library involves a layered architecture that sequentially implements a knowledge database module for collecting and storing domain data, a knowledge system module for establishing a classification and organizational structure that conforms to domain characteristics, a knowledge processing and extraction module for in-depth processing and information extraction of domain data, and a knowledge vectorization module for converting processed domain knowledge into numerical vector representations, forming a complete automated data processing pipeline. It also integrates multi-model intelligent orchestration capabilities, using a model repository for unified registration and version management of various domain-specific models, and utilizing an API gateway to standardize and expose the capabilities of all models, achieving unified scheduling and efficient utilization of model resources. Furthermore, it integrates knowledge application functions, using a domain knowledge graph as the core hub to connect various application modules, establishing a data flow and function call mechanism between modules, forming a collaborative application ecosystem. Finally, it constructs vertical domain intelligent agents, professionally configuring and optimizing parameters based on domain knowledge characteristics and business needs, enabling them to accurately handle complex tasks in specific domains.

[0099] For example, in the field of medical diagnostics, when building a vertical domain vector library, the knowledge base construction module collected 35,000 electronic medical records and medical literature; the knowledge system module established a medical classification system with six dimensions, including disease classification, treatment plans, and drug information; the knowledge processing and extraction module extracted 1.25 million medical entities and relationships; and the knowledge vectorization module used a medical-specific BERT model to generate 768-dimensional medical knowledge vectors, constructing a medical vector library containing 850,000 vectors. When integrating multi-model intelligent orchestration, the model repository uniformly manages 12 medical-specific models, including disease prediction models, image recognition models, and medication recommendation models, and provides 28 API gateways. The system has standardized interfaces, processing an average of 23,000 medical query requests daily. When integrating knowledge application functions, it uses a medical knowledge graph as its core, connecting eight functional modules including intelligent diagnosis, medication recommendation, and medical record analysis. The graph contains 12,000 disease entities, 8,500 symptom entities, and 6,200 drug entities, forming a complete medical application system. In constructing vertical domain intelligent agents, it optimized diagnostic accuracy parameters for disease diagnosis tasks, configured medical professional dictionaries and clinical guideline rules, achieving a diagnostic accuracy rate of 96.8% for 3,000 cases and a drug incompatibility recognition accuracy rate of 98.2% in testing, thus improving the professional level of medical knowledge management.

[0100] The following is combined Figures 2 to 6 Another embodiment of the intelligent knowledge management method of the present invention is described below:

[0101] like Figure 2 As shown, this embodiment includes:

[0102] Data Foundation Layer: The Vector Knowledge Base Management Subsystem serves as the system's data foundation, responsible for the unified governance and vectorization of heterogeneous data. This includes: Multi-dimensional Knowledge System Construction: Supporting the establishment of a multi-dimensional, automatically categorized knowledge organization architecture with parent-child and peer-level relationships; Streamlined Processing and Vectorization: By invoking lower-level AI capabilities, a standardized data processing pipeline is implemented, from layout parsing and structured processing to information extraction, ultimately achieving vectorization. Separation of Management and Application: This subsystem focuses on data management, providing clean, high-quality data services to upper-level applications through standard interfaces.

[0103] Capability Supply Layer: The multi-model intelligent orchestration subsystem serves as the central hub for AI capability scheduling and orchestration. It employs a design that separates resource management from application orchestration. The multi-model management module, acting as the resource layer, uniformly deploys, fine-tunes, evaluates, and API-encapsulates various AI models, achieving standardization and manageability of model capabilities. The agent orchestration module, acting as the application layer, uses a closed-loop architecture to flexibly combine atomic model capabilities and plug-in tools according to dynamic workflows to complete complex, multi-functional tasks.

[0104] Application Service Layer: The knowledge management platform subsystem directly faces users, providing out-of-the-box knowledge applications and domain-specific intelligent agents. This includes: Knowledge Applications: With intelligent question answering as the core entry point, it organically coordinates with five other major functions. Intelligent retrieval, writing, and reading serve as core functional components, mutually invoking each other during execution, supported by underlying data from knowledge graphs and personal knowledge management. Intelligent Agents: It provides five types of domain-specific intelligent agents (question answering, reporting, charting, reviewing, and extraction). The core difference lies in the fact that these agents are not developed independently, but rather rapidly constructed by assembling and invoking the standardized capabilities provided by the lower-level "multi-model intelligent orchestration subsystem."

[0105] Scenario Implementation Layer: Vertical Domain Knowledge Management Applications. This layer embodies the system's scalability. By adopting a model of infrastructure reuse and domain capability assembly on top of the general three-layer subsystem, customized knowledge management applications and intelligent agents for specific domains can be quickly built.

[0106] like Figure 3As shown, the Vector Knowledge Base Management System is geared towards artificial intelligence, providing data governance services for enterprise or institutional data. It forms a data governance foundation based on a large model, responsible for data access, processing, storage, vectorization, and management. The system includes four modules: knowledge base construction, knowledge system, knowledge processing and extraction, and knowledge vectorization. The knowledge base construction module mainly implements three functions: structured data import and database construction, enabling the import and construction of knowledge bases from external databases, supporting both domestic and non-domestic databases such as MySQL, SQL Server, Oracle, and DM Database; Excel database construction, allowing knowledge base construction through uploading and importing between Excel and CSV files; and unstructured data database construction, enabling knowledge base construction through batch uploading of unstructured document data such as single / double-layer PDFs, Word documents, and JPG images.

[0107] The knowledge system module includes knowledge system construction and knowledge classification: The knowledge system construction function can establish a multi-dimensional classification knowledge system, with each classification dimension representing a knowledge organization perspective (e.g., but not limited to: technology field dimension, product line dimension, project affiliation dimension, business process stage dimension, knowledge type dimension, applicable role dimension, geographical region dimension, etc.). For each dimension, it supports establishing its internal hierarchical structure, including parent-child relationships and peer relationships. The knowledge classification function uses a task mode to automatically link (automatically classify) the knowledge system with the knowledge base data. It can classify the same knowledge base data multiple times according to different classification systems, and it can also classify the same knowledge system multiple times according to different knowledge base data. Classification tasks can be published, and the published classification results can provide data support for the knowledge management application module.

[0108] The knowledge processing and extraction module calls the interface of the multi-model intelligent orchestration subsystem to perform three types of processing on the knowledge base data: layout parsing, structure processing, and information extraction. The processing results are stored to support subsequent applications. The layout parsing function primarily understands the physical layout and visual structure of the document, identifies and locates various content areas and their spatial relationships, and uses computer vision (CV) and optical character recognition (OCR) technologies to analyze the position, bounding boxes, and hierarchical nesting relationships of page elements (such as text blocks, tables, images, titles, headers / footers, and lists). The structure processing function converts the text blocks obtained from layout parsing into highly structured data with clearly defined fields, formats, and title hierarchy information, and generates a structured layout element tree / coordinate information. The information extraction function calls the interface of the multi-model intelligent orchestration subsystem to extract deep semantic information from the text. For example, it identifies "product model," "technical parameters," and "operation steps" from technical documents; "people," "organizations," "locations," and "events" and their relationships from news; and extracts "problems" and "sentiment" from user feedback.

[0109] The knowledge vectorization module calls the vectorization model of the multi-model intelligent orchestration subsystem to vectorize knowledge base data, including four functions: format conversion, data slicing, vector knowledge base, and vector verification. The format conversion function primarily transforms heterogeneous raw data in the knowledge base (such as PDF documents, Word files, images, database records, text fragments, etc.) into a standard text format suitable for data slicing and vectorization model processing. The data slicing function, based on the processing capabilities of the vectorization model and downstream application requirements (such as retrieval granularity), segments the converted standard text into appropriately sized, semantically relatively complete fragments. The vector knowledge base transforms text fragments into numerical vectors in a high-dimensional space, capturing the deep semantic information of the text to form a vector knowledge base. The vector verification function includes similarity verification and question-answering source tracing verification, performing quality checks and basic rationality assessments on the generated vectors to ensure effective representation of the semantics of the original text.

[0110] like Figure 4 As shown, the multi-model intelligent orchestration subsystem comprises two modules: multi-model management and agent orchestration. The multi-model management module provides unified control over various AI models, ensuring they are manageable, usable, and traceable, and providing model resources for agent orchestration. It mainly includes four functions: model scheduling management, model training fine-tuning, model evaluation, and model open API. The model scheduling management unit is configured to perform model deployment, resource allocation, operation monitoring, and instance scheduling, ensuring high availability of model services. The model training fine-tuning unit is responsible for model introduction, customized fine-tuning training, and model optimization to generate model versions adapted to specific needs. The model evaluation unit is used to systematically evaluate the model's performance, effectiveness, security, and compliance, generating quantitative evaluation reports. The model open API unit provides standardized interface services, encapsulating and exposing model capabilities to external systems, enabling manageable invocation of model services.

[0111] The intelligent agent orchestration module enables collaborative work among multiple agents, ensuring the efficient completion of complex tasks. The intelligent orchestration module consists of five parts: a knowledge base, intelligent prompts, agents, plugins, and workflow. The knowledge base provides structured data support; intelligent prompts parse task instructions and trigger agent behaviors; plugins extend the capability boundaries of agents, providing atomic-level tool calls; the workflow dynamically orchestrates the collaborative order of agents and plugins using directed graph logic, achieving task decomposition, data flow, and result aggregation; and agent management stores and manages agents constructed through the intelligent orchestration workflow, performing classification, input / output management, parameter control, and operational monitoring. These five parts form a closed-loop architecture, ensuring the efficient collaborative completion of complex tasks.

[0112] like Figure 5As shown, the knowledge management platform subsystem comprises two parts: knowledge application and intelligent agents. The knowledge application part includes six core functions: retrieval, question answering, writing, reading, knowledge graph, and personal knowledge management. Intelligent question answering serves as the core entry point of the entire platform, automatically invoking other functions by recognizing the user's question intent. Intelligent retrieval responds to question-and-answer calls, providing accurate information retrieval and ranking, and supplying materials for writing and reading. Intelligent reading, while analyzing documents, can embed question-and-answer interactions for understanding and calls upon intelligent agents to generate reports stored in the personal knowledge base. Intelligent writing, during the creation process, can initiate retrieval and question-and-answer functions within the system at any time to retrieve knowledge and gain inspiration, generating documents that are stored in the personal knowledge base for accumulation. The knowledge graph and personal knowledge management together form the underlying support. The knowledge graph provides semantic associations and graph recommendations for retrieval and question answering, while personal knowledge management acts as a data center, collecting knowledge produced by various modules and also collecting knowledge directly uploaded by users, processing and vectorizing it before feeding it back to the entire platform, forming a knowledge loop.

[0113] The intelligent agent component is a collection of intelligent agents built upon the multi-model intelligent orchestration subsystem and combined with specific domain business application scenarios. Intelligent agents are mainly divided into five categories: domain question-answering intelligent agents, report generation intelligent agents, chart analysis intelligent agents, document review intelligent agents, and information extraction intelligent agents.

[0114] The five types of intelligent agents rely on the following key technologies for implementation:

[0115] Domain-specific question-answering agents: Based on a Retrieval Augmentation (RAG) architecture, combined with semantic retrieval and generative large models (such as LLM), they achieve accurate querying and semantic reasoning of structured and unstructured knowledge bases. They support multi-turn dialogue management and context awareness, and can integrate domain knowledge graphs to enhance logical consistency.

[0116] Report generation intelligent agent: It adopts multimodal data fusion and template generation technology, integrates text summarization, table generation and data visualization components, and combines controllable text generation technology to ensure content compliance and style consistency.

[0117] Chart analysis-based intelligent agents: Relying on a data-driven analysis engine, they integrate statistical modeling, time series prediction, and computer vision technologies, support automatic feature extraction and visual recommendations, and lower the barrier to entry through natural language interaction.

[0118] Document review intelligent agent: Combining rule engine and deep learning model, it realizes text compliance check, risk identification, version comparison and anomaly detection, and supports multi-dimensional strategy configuration and audit traceability.

[0119] Information extraction intelligent agents: Breaking the limitations of fixed templates, they support users to define extraction fields, relationships, and output formats, achieving accurate conversion of unstructured and semi-structured data into structured results.

[0120] like Figure 6 As shown, taking a specific domain as an example, the vertical domain knowledge management application, based on the general capabilities built upon the aforementioned three core subsystems (vector knowledge base, multi-model orchestration, and knowledge management platform), forms a vertical domain vector library for "vertical domain knowledge management application":

[0121] The core of this subsystem is establishing a data center and automating data processing and vectorization. Key construction components include: Layered construction: Following the data processing flow, four modules are built layer by layer: knowledge database, knowledge system, knowledge processing and extraction, and knowledge vectorization. Heterogeneous compatibility: Access and parsing solutions for structured (DB), semi-structured (Excel / CSV), and unstructured (PDF / Word / image) data are fully considered. Separation of management and application: This subsystem module is responsible for data processing and management, and its output supports the front-end application. Capability invocation: The knowledge processing, extraction, and vectorization modules do not develop underlying AI capabilities themselves, but instead call dedicated model services provided by the "Multi-Model Intelligent Orchestration Subsystem" through standard interfaces, avoiding redundant functional development.

[0122] The core of this multi-model intelligent orchestration subsystem is to enable the output of AI capabilities across the entire system, managing computing power and models downwards and providing intelligent services upwards. Key construction components include: Dual-module drive: Divided into two sub-modules: "Multi-model Management" and "Agent Orchestration." Unified management: Internal and external models are uniformly registered, version-managed, and their lifecycle monitored through a model repository. Standardized interfaces: All model capabilities are exposed through a unified, standardized API gateway, facilitating calls from other subsystems. Separation of orchestration and execution: The agent orchestration module defines the workflow, while the execution engine handles the actual model calls and tool execution.

[0123] The knowledge application function integration subsystem directly provides application functions to users. Using a knowledge graph as a hub, it connects the data flow of various functional modules, creating a synergistic effect. Its main construction contents include: Module interaction design: Using question-and-answer as the core entry point, it clarifies the calling relationships between various modules. Capability reuse: All functional modules do not directly process underlying data, but instead achieve this by calling the AI ​​capabilities of the "multi-model intelligent orchestration subsystem." Knowledge graph-driven: Using the knowledge graph as the core data structure, it connects modules such as retrieval, question-and-answer, and writing, providing deeper semantic association and reasoning capabilities. Unified entry point: Through the "Personal Knowledge Management" module, it provides users with unified knowledge storage, while simultaneously providing user data output to various modules to complete the knowledge loop.

[0124] The Vertical Domain Intelligent Agent Construction subsystem, building upon the infrastructure and general capabilities provided by the first three subsystems, rapidly constructs domain-specific intelligent agents tailored to specific business scenarios. Its main components include: A marketplace model: Employing the concept of intelligent agents, it provides five types of out-of-the-box intelligent agent templates, lowering the barrier to entry. Assembly rather than creation: The capabilities of intelligent agents are not developed from scratch, but rather achieved through orchestrating and combining intelligent agents, tools, and workflows within the "Multi-Model Intelligent Orchestration Subsystem." Domain focus: Each type of intelligent agent is deeply optimized and configured for specific domain tasks (such as question answering, reporting, analysis, and auditing).

[0125] For the purpose of simplicity, the method steps disclosed in the above embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0126] like Figure 7 As shown, the present invention also provides an intelligent knowledge management system, comprising:

[0127] The knowledge base construction module 201 is configured to build a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data.

[0128] The data processing module 202 is configured to process the knowledge data, including layout parsing, structured processing and semantic information extraction, to generate processed knowledge data;

[0129] The vector transformation module 203 is configured to transform the processed knowledge data into vector form and build a vector knowledge base. The transformation process includes format conversion, data slicing and vector generation.

[0130] The knowledge management and optimization module 204 is configured to provide knowledge application functions and domain intelligent agent services based on the vector knowledge base, through dynamic workflow combination of AI model capabilities and plug-in tools, and capture user-derived knowledge to flow back to the vector knowledge base, forming a knowledge closed loop.

[0131] It is worth noting that although only some basic functional modules are disclosed in the embodiments of this invention, it does not mean that the composition of this system is limited to the above-mentioned basic functional modules. On the contrary, what this embodiment intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can arbitrarily add one or more functional modules in combination with existing technology to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. The fact that this embodiment only discloses a few basic functional modules should not be considered as the scope of protection of the claims of this invention being limited to the disclosed basic functional modules. At the same time, for the convenience of description, the above device is described separately according to its functions as various units and modules. Of course, in implementing this invention, the functions of each unit and module can be implemented in one or more software and / or hardware.

[0132] like Figure 8 As shown, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of an intelligent knowledge management method.

[0133] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. For example... Figure 8 The structure shown in this embodiment of the invention includes an electronic device comprising one or more processors 710 and a memory 720; the processors 710 in this electronic device may be one or more. Figure 8 Taking a processor 710 as an example; the memory 720 is used to store one or more programs; the one or more programs are executed by the one or more processors 710, so that the one or more processors 710 implement an intelligent knowledge management method as described in any one of the embodiments of the present invention.

[0134] The electronic device may also include an input device 730 and an output device 740.

[0135] The processor 710, memory 720, input device 730, and output device 740 in this electronic device can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0136] The memory 720 in this electronic device serves as a computer-readable storage medium, capable of storing one or more programs. These programs can be software programs, computer-executable programs, or modules, such as the program instructions / modules corresponding to the intelligent knowledge management method provided in this embodiment of the invention. The processor 710 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 720, thereby implementing the intelligent knowledge management method described in the above embodiment.

[0137] The memory 720 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 720 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 720 may further include memory remotely located relative to the processor 710, which can be connected to the device 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.

[0138] Input device 730 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 740 may include display devices such as a display screen.

[0139] The present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of an intelligent knowledge management method.

[0140] Specifically, the computer storage medium in this embodiment of the invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent knowledge management method, characterized in that, include: Build a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data; The knowledge data is processed, including layout parsing, structure processing, and semantic information extraction, to generate processed knowledge data; The processed knowledge data is transformed into vector form to build a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation. Based on the aforementioned vector knowledge base, AI model capabilities and plug-in tools are combined through a dynamic workflow to provide knowledge application functions and domain intelligent agent services, and user-derived knowledge is captured and fed back to the vector knowledge base to form a knowledge closed loop.

2. The intelligent knowledge management method according to claim 1, characterized in that, Construct a unified knowledge base by importing structured data, semi-structured files, and unstructured documents from heterogeneous data sources, establishing a multi-dimensional classification system with parent-child and peer relationships, and generating classified knowledge data, further including: Import structured data from a relational database, establish a connection with the source database, and extract the data table structure and records; Import data from semi-structured files, parse the file format, and extract structured information; Import data from unstructured documents, identify document types, and extract text content; Establish a multi-dimensional classification system with parent-child and peer-level relationships. The classification system includes classification perspectives based on technical field, product line, project affiliation, business process stage, knowledge type, applicable role, and geographical region. The automatic classification task connects knowledge data with the classification system, generating classified knowledge data.

3. The intelligent knowledge management method according to claim 1, characterized in that, The knowledge data is processed, including layout parsing, structured processing, and semantic information extraction, to generate processed knowledge data, further including: Perform layout analysis to identify the physical layout and visual structure of the document, and locate the content areas and spatial relationships within the document; The text blocks obtained from page layout parsing are converted into structured data with clearly defined fields, formats, and heading hierarchies. Information extraction is performed to extract deep semantic information from text, including entity recognition and relation extraction. The system stores the processing results, including the results of layout parsing, structured processing, and information extraction.

4. The intelligent knowledge management method according to claim 1, characterized in that, The processed knowledge data is transformed into vector form to construct a vector knowledge base. This transformation process includes format conversion, data slicing, and vector generation, and further includes: Perform format conversion to unify the heterogeneous raw data in the knowledge base into a standard text format suitable for data slicing; Data slicing is performed, and standard text is divided into semantically complete text fragments based on the processing capabilities of the vectorization model and the needs of downstream applications. Vector generation is performed, and a vectorization model is used to convert text fragments into vector representations to form a vector knowledge base; Vector validation is performed, and the quality of the generated vectors is evaluated through similarity validation and question-answering source tracing validation.

5. The intelligent knowledge management method according to claim 1, characterized in that, Provides knowledge application functions and domain-specific intelligent agent services, further including: It provides intelligent question-answering functionality, which identifies the user's question intent and automatically retrieves other functions; It provides intelligent search functionality to retrieve information for question-and-answer sessions and to provide materials for writing and reading. It provides intelligent writing and intelligent reading functions, and calls retrieval and question-and-answer functions during the execution process, and stores the generated content in the knowledge base; It provides knowledge graph and personal knowledge management functions, provides underlying data support, and collects the generated knowledge; Construct domain-specific intelligent agent services, including domain-specific question-answering, report generation, chart analysis, document review, and information extraction intelligent agents.

6. The intelligent knowledge management method according to claim 5, characterized in that, Building domain-specific intelligent agent services includes: Construct a domain-specific question-answering intelligent agent based on a retrieval-enhanced generative architecture, combined with semantic retrieval and a generative large model; Construct a report generation intelligent agent that integrates text summarization, table generation, and data visualization components based on multimodal data fusion and template generation; Construct a chart analysis-type intelligent agent based on a data-driven analysis engine, integrating statistical modeling, time series forecasting, and computer vision technologies; Construct a document review-type intelligent agent, combining a rule engine and a deep learning model, to achieve text compliance checks, risk identification, and version comparison; Construct an information extraction intelligent agent to support users in defining extraction fields, relationships, and output formats, thereby transforming unstructured and semi-structured data into structured results.

7. The intelligent knowledge management method according to claim 5, characterized in that, The method also includes the step of building vertical domain knowledge management applications: Build a vertical domain vector library, and achieve automated data processing by constructing a knowledge library, knowledge system, knowledge processing and extraction, and knowledge vectorization in a layered manner; It integrates intelligent orchestration of multiple models, and manages model capabilities in a unified manner through a model repository and API gateway. All model capabilities are exposed to the outside world through a unified and standardized API gateway. Integrate knowledge application functions, using knowledge graphs as a hub to create a synergistic effect; Build vertical domain intelligent agents and perform in-depth optimization and configuration for specific domain tasks.

8. An intelligent knowledge management system, characterized in that, include: The knowledge base construction module is configured to build a unified knowledge base, import structured data, semi-structured files and unstructured documents from heterogeneous data sources, establish a multi-dimensional classification system with parent-child and peer relationships, and generate classified knowledge data. The data processing module is configured to process the knowledge data, including layout parsing, structure processing and semantic information extraction, to generate processed knowledge data; The vector transformation module is configured to convert processed knowledge data into vector form and build a vector knowledge base. The transformation process includes format conversion, data slicing, and vector generation. The knowledge management and optimization module is configured to provide knowledge application functions and domain intelligent agent services based on the vector knowledge base, through dynamic workflow combination of AI model capabilities and plug-in tools, and capture user-derived knowledge to flow back to the vector knowledge base, forming a knowledge closed loop.

9. An electronic device, characterized in that, include: The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method according to any one of claims 1 to 7.

Images