An AI-based enterprise knowledge base construction system and method
By building an AI-based enterprise knowledge base system, and utilizing technologies such as semantic vector matching and the construction skills of rights and responsibilities links, the system solves the problems of inaccurate matching of rights and responsibilities clauses, loose relationships, and insufficient identification of confidential fragments in enterprise knowledge bases, thus achieving more efficient and secure knowledge management and storage.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN TONGWEI ELECTRONICS CO LTD
- Filing Date
- 2026-06-15
- Publication Date
- 2026-07-31
AI Technical Summary
Existing technologies for building enterprise knowledge bases suffer from problems such as inaccurate matching of rights and responsibilities, loose relationships, insufficient identification of confidential information, and non-compliant storage architecture, resulting in high processing costs and insufficient security and compliance.
By employing technologies such as semantic vector matching and parameter set verification of rights and responsibilities clauses, rights and responsibilities link construction skills, classified fragment identification and hierarchical storage, and AI compliance verification assistant, we can achieve accurate matching of rights and responsibilities clauses, tight construction of related relationships, accurate identification of classified fragments, and compliant storage.
It improves the semantic accuracy and automated processing efficiency of knowledge collection, builds a more complete and tightly linked chain of rights and responsibilities, realizes the isolated storage of classified and non-classified knowledge, and ensures the compliance and security of the knowledge base throughout its entire lifecycle.
Smart Images
Figure CN122489680A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of knowledge base technology, and in particular to an AI-based enterprise knowledge base construction system and method. Background Technology
[0002] The knowledge base technology field encompasses multiple technical branches, including the full lifecycle management of knowledge assets, structured and unstructured processing of knowledge data, knowledge retrieval and reuse, and knowledge association and iteration. The core of this technology field revolves around the knowledge resources generated during the production and operation processes of various entities. Through standardized technical paths, it achieves the collection, classification, integration, and transfer of knowledge, covering the entire process of collection, governance, storage, and application of various knowledge carriers such as institutional documents, business experience, project results, and industry materials. This provides underlying technical support for the accumulation, transmission, and standardized management of knowledge assets, and is a core technology field for realizing the management of knowledge assets during digital transformation.
[0003] Among them, the AI-based enterprise knowledge base construction system and method refers to the technical solutions related to knowledge base construction for enterprise operation and management scenarios. The technical aspects addressed by this solution include: standardized collection of multi-source heterogeneous knowledge data, accurate identification of knowledge entities, automatic construction of knowledge relationships, hierarchical classification of knowledge content, establishment of knowledge storage structures, and generation of knowledge retrieval indexes. This solution typically employs natural language processing technology for word segmentation and entity extraction of unstructured text, knowledge graph technology for constructing knowledge nodes and related edges, text classification technology for category matching of knowledge content, vector databases for storing and indexing knowledge features, and rule-based matching logic for compliance verification and update triggering of knowledge content.
[0004] Existing technologies for processing corporate policy documents employ generalized text processing, relying solely on word segmentation and entity extraction without designing specific matching and filtering logic for rights and responsibilities clauses. This leads to the mixing of irrelevant content into the aggregated knowledge, increasing processing costs. For example, non-rights and responsibilities content is aggregated together, requiring manual screening. Furthermore, the reliance on knowledge graphs to build relationships fails to focus on cross-matching of rights and responsibilities parameters, resulting in loose connections that cannot reconstruct the entire logical chain. When performing security level matching on complete documents, the technology fails to separate and identify confidential segments, either hindering the reuse of non-confidential content or posing a risk of leakage. Finally, the fixed storage architecture lacks parameter linkage, making it impossible to adjust promptly to parameter changes, resulting in insufficient compliance. Summary of the Invention
[0005] The main objective of this invention is to provide an AI-based enterprise knowledge base construction system and method, which can effectively solve the problems mentioned in the background.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0007] An AI-based enterprise knowledge base construction system, the system being configured as follows:
[0008] Clause traversal and matching module: Obtain the full text of the enterprise's policy documents, traverse them line by line, generate the benchmark rules for matching rights and responsibilities clauses based on prompt words, decompose each text into independent semantic units and generate semantic vectors, and generate feature vectors through quantization. Perform vector matching verification with the parameter set associated with rights and responsibilities clauses to filter content that meets the rules and generate the core entries of rights and responsibilities clauses.
[0009] Link association comparison module: calls the core items of the rights and responsibilities clauses, extracts the associated parameter items, calls the rights and responsibilities link construction skill to cross-match and verify the parameters of all items, marks the association relationship, anchors the link nodes and builds the rights and responsibilities knowledge graph, matches the enterprise system rights and responsibilities clause collection library rules, and generates the full-link rights and responsibilities collection link;
[0010] The classified fragment identification module calls the full-link responsibility collection link, splits text fragments, performs vector similarity matching verification based on AI classified identification prompt words and classified classification parameter set, compares with classified judgment rules, filters fragments that meet the rules, binds identifiers and parameters, and generates classified fragment control entries.
[0011] Hierarchical storage mapping module: calls classified fragment control entries, matches the classification level and isolation subdomain, stores fragments in a targeted manner, generates index units based on vector database, binds indexes with collection links and responsibility knowledge graphs, and establishes a hierarchical storage architecture for the knowledge base;
[0012] Parameter linkage and adaptation module: It calls the hierarchical storage architecture of the knowledge base, integrates content, calls the AI compliance verification assistant to verify parameters, compares and builds benchmark rules, adjusts node configuration through parameter adaptation skills, and generates a full-chain compliance knowledge base for enterprises.
[0013] Preferably, in the clause traversal and matching module, the line-by-line splitting traversal is to split the entire text of the enterprise system document according to sentence semantics, decompose a single text into an independent semantic unit and generate a semantic vector, compare the semantic vector with the set of parameters associated with the rights and responsibilities clauses one by one, and determine that it conforms to the matching benchmark rules when the semantic vector matches at least two parameter items.
[0014] Preferably, in the link association comparison module, the extracted association parameters are specifically three independent parameters: pre-triggering conditions, parallel execution items, and post-accountability constraints. The marked association relationships are divided into three categories: pre-association, parallel association, and post-association. The association relationships are synchronously mapped to the node edges of the rights and responsibilities knowledge graph.
[0015] Preferably, the skill is constructed through the responsibility chain, comparing the pre-trigger condition parameters of a single item with the semantic units of the execution results of the remaining items, comparing the parameters of parallel execution items with the semantic units of the synchronous execution of the remaining items, and marking the corresponding association when the semantics are completely consistent and updating the association of the knowledge graph nodes.
[0016] Preferably, in the classified segment recognition module, the text segment is split into individual semantic units by paragraph punctuation. After the semantic units are vectorized, they are compared with the classification level parameter set for similarity. When the number of matching parameter items reaches two or more, it is determined that it meets the AI classification judgment rules.
[0017] Preferably, the binding identifier and parameters are to assign a unique string identifier to the fragment that meets the confidentiality determination rules, and to associate and bind the identifier with the fragment's confidentiality level, the fragment's confidentiality boundary, and the fragment's scope of knowledge parameters one by one. The parameters are synchronized to the knowledge graph and vector database.
[0018] Preferably, in the hierarchical storage mapping module, the isolated subdomains are set according to the security level, and a one-way data control link is used between each isolated subdomain. Only the high-security subdomain is allowed to initiate parameter verification requests to the low-security subdomain. The data stored in the subdomains are all stored in the vector database in vector form.
[0019] Preferably, in the parameter linkage adaptation module, the verification parameters are specifically four types of parameters: verification security level parameters, knowledge scope parameters, associated link node parameters, and storage node permission parameters. The verification process is automatically executed by the AI compliance verification assistant.
[0020] Preferably, the adjustment node is configured such that when the current value of the parameter is inconsistent with the standard value of the construction benchmark rule parameter, the permissions and associated link configuration of the corresponding storage node are adjusted synchronously through the parameter adaptation skill, and the parameter information in the index unit, knowledge graph and vector database is updated synchronously.
[0021] In addition, the present invention also provides an AI-based method for constructing an enterprise knowledge base, the method comprising the following steps:
[0022] S1. Obtain the full text of the enterprise's policy documents, split and traverse them line by line, generate a benchmark rule for matching rights and responsibilities clauses based on prompt words, decompose a single text into an independent semantic unit and generate a semantic vector, and generate a feature vector from the quantization, perform vector matching and verification with the parameter set associated with the rights and responsibilities clauses to filter out content that meets the rules, and generate the core items of the rights and responsibilities clauses.
[0023] S2. Call the core entries of the rights and responsibilities clauses, extract the related parameter items within the entries, call the rights and responsibilities link construction skill to perform parameter cross-matching and verification, mark the relationship, anchor the link nodes and build a rights and responsibilities knowledge graph, and generate the full-link rights and responsibilities collection link.
[0024] S3. Call the full-link responsibility collection link, split the corresponding content of the link into continuous text fragment units, perform vector similarity matching verification on the text fragment units based on AI classified identification prompt words and classified classification parameter set, compare with the classified judgment benchmark rules, filter fragments that meet the rules, bind the fragment unique identifier and corresponding parameters, and generate classified fragment control entries.
[0025] S4. Call the classified fragment control entries, match the classification level and isolated subdomains, store fragments in a targeted manner, generate an index based on the vector database, bind the index with the collection link and knowledge graph, and establish a hierarchical storage architecture for the knowledge base.
[0026] S5 invokes the hierarchical storage architecture of the knowledge base, verifies all parameters through the AI compliance verification assistant, compares them with the benchmark rules, adjusts the node configuration through the parameter adaptation skill, and generates an enterprise full-chain compliance knowledge base.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] This invention sets matching rules based on prompt words and performs text matching verification through semantic vectors and vector similarity. It acquires the full text of enterprise policy documents, splits and traverses it line by line, matches and verifies individual texts with the parameter sets associated with rights and responsibilities clauses, and compares them with the matching benchmark rules. Content that meets the rules is selected to generate core entries, significantly improving the semantic accuracy of knowledge collection and the efficiency of AI automated processing. It completes cross-matching verification of all entry parameters through a rights and responsibilities link construction skill, marks the relationships to anchor link nodes, and constructs a rights and responsibilities knowledge graph, building a more complete and tightly linked rights and responsibilities logical link. It splits text fragments and performs matching verification of the confidentiality level parameter set based on AI-based confidentiality identification prompt words and vector comparison, enabling more accurate and fine-grained identification of confidential fragments. It matches the confidentiality level and isolates subdomains for targeted storage, generates an index based on the vector database, and binds links and knowledge graphs, achieving isolation and traceability of confidential and non-confidential knowledge, improving knowledge retrieval and storage efficiency. Through AI... The compliance verification assistant and parameter adaptation skill verify parameters, compare benchmark rules, and adjust node configurations to achieve real-time parameter linkage adaptation and automatic compliance verification, ensuring compliance throughout the knowledge base lifecycle and further improving the standardization and security of knowledge management. Attached Figure Description
[0029] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation
[0030] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0031] Example 1, as Figure 1 As shown, an AI-based enterprise knowledge base construction system is configured as follows:
[0032] Clause traversal and matching module: Obtain the full text of the enterprise's policy documents, traverse them line by line, generate the benchmark rules for matching rights and responsibilities clauses based on prompt words, decompose each text into independent semantic units and generate semantic vectors, and generate feature vectors through quantization. Perform vector matching verification with the parameter set associated with rights and responsibilities clauses to filter content that meets the rules and generate the core entries of rights and responsibilities clauses.
[0033] Link association comparison module: calls the core items of the rights and responsibilities clauses, extracts the associated parameter items, calls the rights and responsibilities link construction skill to cross-match and verify the parameters of all items, marks the association relationship, anchors the link nodes and builds the rights and responsibilities knowledge graph, matches the enterprise system rights and responsibilities clause collection library rules, and generates the full-link rights and responsibilities collection link;
[0034] The classified fragment identification module calls the full-link responsibility collection link, splits text fragments, performs vector similarity matching verification based on AI classified identification prompt words and classified classification parameter set, compares with classified judgment rules, filters fragments that meet the rules, binds identifiers and parameters, and generates classified fragment control entries.
[0035] Hierarchical storage mapping module: calls classified fragment control entries, matches the classification level and isolation subdomain, stores fragments in a targeted manner, generates index units based on vector database, binds indexes with collection links and responsibility knowledge graphs, and establishes a hierarchical storage architecture for the knowledge base;
[0036] Parameter linkage and adaptation module: It calls the hierarchical storage architecture of the knowledge base, integrates content, calls the AI compliance verification assistant to verify parameters, compares and builds benchmark rules, adjusts node configuration through parameter adaptation skills, and generates a full-chain compliance knowledge base for enterprises.
[0037] Specifically, a medium-sized manufacturing enterprise is selected as the application scenario. Assume this enterprise has 12 enterprise policy documents, including management systems, job descriptions, and business process specifications, totaling 8600 words. First, the clause traversal and matching module is activated to obtain the full text of these 12 enterprise policy documents. The documents are then split and traversed line by line. Each split clause is matched and verified against a preset set of parameters related to rights and responsibilities. This parameter set includes three parameters: pre-triggered conditions, parallel execution items, and post-accountability constraints. The matching results are then compared with preset matching benchmark rules to filter out text content that meets the rules, generating core clauses of rights and responsibilities. These core clauses are then invoked, and the link association comparison module is activated to extract the associated parameter items within each clause. Cross-matching verification is performed on all clause parameters to mark the relationships between clauses, anchor link nodes, and match the enterprise's preset enterprise policy rights and responsibilities clause collection rules to generate a full-link rights and responsibilities collection link. This full-link link is then invoked. The responsibility and authority aggregation link initiates the classified fragment identification module, splits the corresponding content of the link into continuous text fragment units, and performs vector similarity matching verification on each text fragment unit based on AI classified identification prompts and classified classification parameter sets. It compares the data against the classified judgment rules, filters out text fragments that meet the rules, and binds an identifier and corresponding parameters to each compliant fragment, generating a classified fragment control entry. Calling this classified fragment control entry initiates the hierarchical storage mapping module, matches the classification level of each entry with the preset isolation subdomain, and stores the fragments in the corresponding isolation subdomain, generating an index unit for each fragment. The index unit is uniquely bound to the original aggregation link, establishing a hierarchical storage architecture for the knowledge base. Calling this hierarchical storage architecture initiates the parameter linkage adaptation module, integrates all entries and link content within the architecture, verifies various parameters within the architecture, compares them with the enterprise's preset knowledge base construction benchmark rules, adjusts the corresponding node configurations, and generates an enterprise full-link compliance knowledge base.
[0038] In the clause traversal and matching module, the line-by-line splitting traversal is to split the entire text of the enterprise system document according to the sentence semantics, decompose each text into an independent semantic unit and generate a semantic vector, and compare the semantic vector with the parameter set associated with the rights and responsibilities clauses one by one. When the semantic vector matches at least two parameter items, it is determined to meet the matching benchmark rules.
[0039] In the link association comparison module, the extracted association parameters are specifically three independent parameters: pre-trigger conditions, parallel execution items, and post-accountability constraints. The marked association relationships are divided into three categories: pre-association, parallel association, and post-association. The association relationships are synchronously mapped to the node edges of the rights and responsibilities knowledge graph.
[0040] Skills are constructed through the rights and responsibilities chain. The pre-trigger condition parameters of a single item are compared with the semantic units of the execution results of the other items. The parameters of parallel execution items are compared with the semantic units of the synchronous execution of the other items. When the semantics are completely consistent, the corresponding relationship is marked and the knowledge graph node association is updated.
[0041] Furthermore, when the clause traversal and matching module performs line-by-line splitting traversal, it splits the entire text of the enterprise policy document according to sentence semantics, decomposes each text into independent semantic units, and generates semantic vectors. For example, "When an employee fails to submit approval according to the process, their process operation should be suspended, and the department head should be held accountable" is split into three independent semantic units: "Employee failed to submit approval according to the process," "Suspend their process operation," and "The department head should be held accountable." Then, each semantic unit is compared one by one with the three parameters in the power and responsibility clause association parameter set: pre-trigger conditions, parallel execution items, and post-accountability constraints. The matching benchmark rule is set so that a semantic unit is judged to meet the rule if it matches at least two parameters. For example, "Employee failed to submit approval according to the process" matches the pre-trigger conditions and post-accountability constraints, and is judged to meet the matching benchmark rule. The link association comparison module extracts the association parameter items specifically as pre-trigger conditions, parallel execution items, and post-accountability constraints. The accountability constraint has three independent parameters. The labeled associations are divided into three categories: pre-association, parallel association, and post-association. The associations are synchronously mapped to the nodes of the responsibility knowledge graph. The parameter cross-matching verification specifically compares the pre-triggering condition parameter of a single item with the semantic unit of the execution result of the other items, and compares the parallel execution parameter with the semantic unit of the synchronous execution of the other items. When the semantics are completely consistent, the corresponding association is marked and the knowledge graph node association is updated. For example, the pre-triggering condition parameter of item A is "employee did not submit approval according to the process", and the semantic unit of the execution result of item B is "employee did not submit approval according to the process". The two are semantically completely consistent, so item A and item B are marked as pre-association. The parallel execution parameter of item C is "department head review", and the semantic unit of the synchronous execution of item D is "department head review". The two are semantically completely consistent, so item C and item D are marked as parallel association.
[0042] In the classified fragment recognition module, text fragments are split into single semantic units by paragraph punctuation. After the semantic units are vectorized, they are compared with the classification level parameter set. When the number of matching parameter items reaches two or more, it is determined that it meets the AI classification judgment rules.
[0043] The binding identifier and parameters assign a unique string identifier to each fragment that meets the classification rules. The identifier is then linked to the fragment's classification level, the fragment's classification boundary, and the fragment's scope of knowledge. The parameters are synchronized to the knowledge graph and vector database.
[0044] In the hierarchical storage mapping module, isolated subdomains are set according to security level. A one-way data control link is used between each isolated subdomain. Only high-security subdomains are allowed to send parameter verification requests to low-security subdomains. All data stored in the subdomains is stored in the vector database in vector form.
[0045] Furthermore, when the classified fragment identification module performs the text fragment splitting operation, it splits the content corresponding to the entire link of responsibility and authority collection according to paragraph punctuation. Each text fragment independently carries a single classified semantic. For example, "The core technical parameters of the classified project are XXX, which can only be viewed by core personnel of the technical department, and the validity period is 1 year" is split into two text fragments by a period. Each fragment is a single semantic unit. Then, each semantic unit is compared with five parameters in the classified classification parameter set: fragment classification level, fragment classification boundary, fragment knowledge scope, fragment validity period, and fragment responsibility subject. The classification judgment rule is set to determine compliance if two or more parameters are matched. For example, if one fragment matches the fragment classification level and fragment knowledge scope parameters, it is determined to comply with the classification judgment rule. The binding identifier and parameter operation specifically assigns a unique string identifier to fragments that comply with the classification judgment rule, such as assigning "SM2". The identifier "0240501001" is used as an identifier, and this identifier is associated with three parameters: the security classification level (confidential), the security boundary of the fragment (core technical parameters), and the scope of knowledge of the fragment (core personnel of the technical department). The isolated subdomains in the hierarchical storage mapping module are divided into three independent levels according to the security classification level: top secret, confidential, and secret. Each level corresponds to a unique security classification identifier: the identifier for top secret is "JM001", for confidential it is "JM002", and for secret it is "JM003". A one-way data control link is used between the isolated subdomains. The link control rule is set to allow only the high-security subdomain to initiate parameter verification requests to the low-security subdomain. The data stored in the subdomains is stored in the vector database in vector form. For example, the top secret subdomain can initiate parameter verification requests to the confidential and secret subdomains, and the confidential subdomain can initiate parameter verification requests to the secret subdomain. It is prohibited for the low-security subdomain to transmit any data to the high-security subdomain.
[0046] In the parameter linkage adaptation module, the verification parameters are specifically divided into four categories: security level parameters, scope of knowledge parameters, associated link node parameters, and storage node permission parameters. The verification process is automatically executed by the AI compliance verification assistant.
[0047] When the current value of the node configuration parameter is inconsistent with the standard value of the baseline rule parameter, the corresponding storage node permissions and associated link configuration are adjusted synchronously through the parameter adaptation skill, and the parameter information in the index unit, knowledge graph and vector database is updated synchronously.
[0048] Furthermore, when the parameter linkage adaptation module performs parameter verification operations, it specifically verifies four types of parameters: security level parameters, scope of knowledge parameters, associated link node parameters, and storage node permission parameters. The verification process is handled by AI. The compliance verification assistant executes automatically, setting standard values for various parameters as the baseline rule parameters. The standard value for the security level parameter corresponds to the security level identifier of the isolated subdomain; the standard value for the scope of knowledge parameter is the personnel list for each department; the standard value for the associated link node parameter is the preset link node association rule; and the standard value for the storage node permission parameter is the access permission list corresponding to each security level. Specifically, when the current parameter value is inconsistent with the standard value of the baseline rule parameters, the corresponding storage node permissions and associated link configurations are adjusted synchronously using the parameter adaptation skill. The parameter information in the index unit, knowledge graph, and vector database is also updated synchronously. For example, if the verification finds that the current value of the scope of knowledge parameter for a classified segment is "all personnel in the technical department," which is inconsistent with the standard value of the baseline rule parameter "core personnel in the technical department," the access permissions of the storage node corresponding to that segment are adjusted synchronously, granting access only to core personnel in the technical department. The configuration of the associated link nodes corresponding to that segment is adjusted, deleting link associations related to non-core personnel in the technical department. Simultaneously, the scope of knowledge parameter information in the index unit of that segment is updated from "all personnel in the technical department" to "core personnel in the technical department," ensuring that the parameters are consistent with the baseline rule.
[0049] In addition, the present invention also provides an AI-based method for constructing an enterprise knowledge base, the method comprising the following steps:
[0050] S1. Obtain the full text of the enterprise's policy documents, split and traverse them line by line, generate a benchmark rule for matching rights and responsibilities clauses based on prompt words, decompose a single text into an independent semantic unit and generate a semantic vector, and generate a feature vector from the quantization, perform vector matching and verification with the parameter set associated with the rights and responsibilities clauses to filter out content that meets the rules, and generate the core items of the rights and responsibilities clauses.
[0051] S2. Call the core entries of the rights and responsibilities clauses, extract the related parameter items within the entries, call the rights and responsibilities link construction skill to perform parameter cross-matching and verification, mark the relationship, anchor the link nodes and build a rights and responsibilities knowledge graph, and generate the full-link rights and responsibilities collection link.
[0052] S3. Call the full-link responsibility collection link, split the corresponding content of the link into continuous text fragment units, perform vector similarity matching verification on the text fragment units based on AI classified identification prompt words and classified classification parameter set, compare with the classified judgment benchmark rules, filter fragments that meet the rules, bind the fragment unique identifier and corresponding parameters, and generate classified fragment control entries.
[0053] S4. Call the classified fragment control entries, match the classification level and isolated subdomains, store fragments in a targeted manner, generate an index based on the vector database, bind the index with the collection link and knowledge graph, and establish a hierarchical storage architecture for the knowledge base.
[0054] S5 invokes the hierarchical storage architecture of the knowledge base, verifies all parameters through the AI compliance verification assistant, compares them with the benchmark rules, adjusts the node configuration through the parameter adaptation skill, and generates an enterprise full-chain compliance knowledge base.
[0055] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. An AI-based enterprise knowledge base construction system, the system being configured as follows: Clause Traversal and Matching Module: Obtain the full text of the enterprise's policy documents, traverse them line by line, generate a baseline rule for matching rights and responsibilities clauses based on prompt words, decompose a single text into an independent semantic unit and generate a semantic vector, generate a feature vector from the quantization, perform vector matching verification with the parameter set associated with the rights and responsibilities clauses, filter out content that meets the rules, and generate the core items of the rights and responsibilities clauses. Link association comparison module: calls the core items of the rights and responsibilities clauses, extracts the associated parameter items, calls the rights and responsibilities link construction skill to cross-match and verify the parameters of all items, marks the association relationship, anchors the link nodes and builds the rights and responsibilities knowledge graph, matches the enterprise system rights and responsibilities clause collection library rules, and generates the full-link rights and responsibilities collection link; The classified fragment identification module calls the full-link responsibility collection link, splits text fragments, performs vector similarity matching verification based on AI classified identification prompt words and classified classification parameter set, compares with classified judgment rules, filters fragments that meet the rules, binds identifiers and parameters, and generates classified fragment control entries. Hierarchical storage mapping module: calls classified fragment control entries, matches the classification level and isolation subdomain, stores fragments in a targeted manner, generates index units based on vector database, binds indexes with collection links and responsibility knowledge graphs, and establishes a hierarchical storage architecture for the knowledge base; Parameter linkage and adaptation module: It calls the hierarchical storage architecture of the knowledge base, integrates content, calls the AI compliance verification assistant to verify parameters, compares and builds benchmark rules, adjusts node configuration through parameter adaptation skills, and generates a full-chain compliance knowledge base for enterprises.
2. The enterprise knowledge base building system of claim 1, wherein: In the clause traversal and matching module, the line-by-line splitting traversal is to split the entire text of the enterprise system document according to the semantics of the sentences, decompose each text into an independent semantic unit and generate a semantic vector, and compare the semantic vector with the parameter set associated with the rights and responsibilities clauses one by one. When the semantic vector matches at least two parameter items, it is determined to meet the matching benchmark rules.
3. The enterprise knowledge base building system of claim 1, wherein: In the link association comparison module, the extracted association parameters are specifically three independent parameters: pre-trigger conditions, parallel execution items, and post-accountability constraints. The marked association relationships are divided into three categories: pre-association, parallel association, and post-association. The association relationships are synchronously mapped to the node edges of the rights and responsibilities knowledge graph.
4. The enterprise knowledge base building system of claim 1, wherein: The skill is constructed through the responsibility chain. The pre-trigger condition parameters of a single item are compared with the semantic units of the execution results of the other items. The parameters of parallel execution items are compared with the semantic units of the synchronous execution of the other items. When the semantics are completely consistent, the corresponding relationship is marked and the knowledge graph node association is updated.
5. The enterprise knowledge base building system of claim 1, wherein: In the classified segment recognition module, the text segment is split into individual semantic units by paragraph punctuation. After the semantic units are vectorized, they are compared with the classification level parameter set. When the number of matching parameter items reaches two or more, it is determined to meet the AI classification judgment rules.
6. The enterprise knowledge base building system of claim 1, wherein: The binding identifier and parameters are used to assign a unique string identifier to fragments that meet the classification rules. The identifier is associated with the fragment's classification level, the fragment's classification boundary, and the fragment's scope of knowledge. The parameters are synchronized to the knowledge graph and vector database.
7. The enterprise knowledge base building system of claim 1, wherein: In the hierarchical storage mapping module, isolated subdomains are set according to security level. A one-way data control link is used between each isolated subdomain. Only high-security subdomains are allowed to initiate parameter verification requests to low-security subdomains. Subdomain storage data is stored in vector database in vector form.
8. The enterprise knowledge base building system of claim 1, wherein: In the parameter linkage adaptation module, the verification parameters are specifically divided into four categories: security level parameters, knowledge scope parameters, associated link node parameters, and storage node permission parameters. The verification process is automatically executed by the AI compliance verification assistant.
9. The enterprise knowledge base building system of claim 1, wherein: When the current value of the parameter is inconsistent with the standard value of the construction benchmark rule parameter, the adjustment node is configured to synchronously adjust the permissions and associated link configuration of the corresponding storage node through the parameter adaptation skill, and synchronously update the parameter information in the index unit, knowledge graph and vector database. 10.A method for constructing an AI-based enterprise knowledge base, the method comprising: The method is used in the enterprise knowledge base construction system according to any one of claims 1-9, and includes the following steps: S1. Obtain the full text of the enterprise's policy documents, split and traverse them line by line, generate a benchmark rule for matching rights and responsibilities clauses based on prompt words, decompose a single text into an independent semantic unit and generate a semantic vector, and generate a feature vector from the quantization, perform vector matching and verification with the parameter set associated with the rights and responsibilities clauses to filter out content that meets the rules, and generate the core items of the rights and responsibilities clauses. S2. Call the core entries of the rights and responsibilities clauses, extract the related parameter items within the entries, call the rights and responsibilities link construction skill to perform parameter cross-matching and verification, mark the relationship, anchor the link nodes and build a rights and responsibilities knowledge graph, and generate the full-link rights and responsibilities collection link. S3. Call the full-link responsibility collection link, split the corresponding content of the link into continuous text fragment units, perform vector similarity matching verification on the text fragment units based on AI classified identification prompt words and classified classification parameter set, compare with the classified judgment benchmark rules, filter fragments that meet the rules, bind the fragment unique identifier and corresponding parameters, and generate classified fragment control entries. S4. Call the classified fragment control entries, match the classification level and isolated subdomains, store fragments in a targeted manner, generate an index based on the vector database, bind the index with the collection link and knowledge graph, and establish a hierarchical storage architecture for the knowledge base. S5 invokes the hierarchical storage architecture of the knowledge base, verifies all parameters through the AI compliance verification assistant, compares them with the benchmark rules, adjusts the node configuration through the parameter adaptation skill, and generates an enterprise full-chain compliance knowledge base.