Search

CN-122021831-A - Distributed knowledge graph construction and encryption storage method based on large model

CN122021831ACN 122021831 ACN122021831 ACN 122021831ACN-122021831-A

Abstract

The invention discloses a large-model-based distributed knowledge graph construction and encryption storage method, and aims to solve the problems of uncontrollable knowledge graph construction quality, low automation degree and insufficient distributed storage safety in the prior art. The method comprises the steps of firstly decomposing an overall construction task, distributing the overall construction task to a plurality of large model intelligent bodies with different capability advantages to perform distributed knowledge extraction, then mutually checking and arbitrating the extracted preliminary knowledge segments through a multi-round cross-validation and consensus achievement mechanism, effectively inhibiting the large model from 'illusion', screening out high-credibility knowledge, and finally performing fine-grained dynamic encryption storage on the high-quality sub-spectrums generated by fusion, namely dynamically generating keys according to the security levels of entities and relations and the structural context thereof in the spectrums, managing the key spectrums with isomorphic keys, and finally performing distributed storage on encrypted data blocks.

Inventors

  • LI XIAOZHAO

Assignees

  • 河北飞驰智芯科技有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (10)

  1. 1. The distributed knowledge graph construction and encryption storage method based on the large model is characterized by comprising the following steps of: Step S1, task planning and agent allocation, namely receiving a knowledge graph construction request, decomposing an overall construction task into a plurality of knowledge extraction subtasks based on a predefined ontology mode, and allocating the subtasks to a plurality of large model agents with corresponding high-priority energy in a distributed network according to a pre-constructed large model agent capability image; s2, carrying out entity identification, relation extraction and attribute filling operations from a designated unstructured or semi-structured data source by each large-model intelligent agent to generate a structured preliminary knowledge segment; step S3, multiple rounds of cross-validation and consensus are achieved, namely, multiple rounds of cross-validation are initiated in a distributed intelligent agent network aiming at the preliminary knowledge segments, the knowledge segments with association or conflict are distributed to at least one other large model intelligent agent for validation, and the knowledge segments with high credibility are screened out based on the confidence and consistency of the validation results to achieve consensus; S4, carrying out logic fusion on the high-credibility knowledge segments which reach consensus, eliminating conflict and forming an internally consistent high-quality quantum spectrum; And S5, carrying out fine-granularity dynamic encryption storage, namely carrying out security level assessment on each entity and relation in the sub-map, dynamically generating an encryption key based on the assessment result and the structural context of the encryption key in the sub-map, carrying out fine-granularity encryption, and finally storing the encrypted data blocks in a plurality of storage nodes in a distributed mode.
  2. 2. The large model-based distributed knowledge graph construction and encryption storage method according to claim 1, wherein the decomposing the overall construction task based on the predefined ontology schema in the step S1 specifically includes: and according to the entity type and the relation type defined in the ontology mode, vertically dividing the construction task according to the entity type and/or horizontally dividing the construction task according to the data source to form the knowledge extraction subtask.
  3. 3. The large-model-based distributed knowledge graph construction and encryption storage method according to claim 1, wherein the multi-round cross validation and consensus achievement in step S3 specifically includes: Setting a confidence threshold; When the confidence coefficient average value of a plurality of verification results aiming at the same knowledge segment is larger than or equal to the confidence coefficient threshold value and the results are consistent, consensus is achieved; When the verification results are inconsistent or the confidence average is below the threshold, the dispute knowledge piece is submitted to a large model agent designated as an arbitrator for final arbitration.
  4. 4. The large model-based distributed knowledge graph construction and encryption storage method according to claim 3, wherein the method further comprises, after step S3 and before step S4: and recording the unqualified dispute knowledge segments and the judging process and result thereof, and using the dispute knowledge segments as feedback data for updating the large model intelligent capability image.
  5. 5. The large-model-based distributed knowledge graph construction and encryption storage method according to claim 1, wherein the step S5 of dynamically generating an encryption key based on the evaluation result and the structural context thereof in the graph is specifically as follows: And a key derivation algorithm is adopted, and a unique identifier of an entity or a relationship, a security level label thereof and an identifier combination of a direct association neighbor are taken as input parameters to derive and generate a unique key.
  6. 6. The large model-based distributed knowledge graph construction and encryption storage method according to claim 5, further comprising: And constructing a key map which is isomorphic with the knowledge map and is used for storing the encryption key or key index corresponding to each entity node and the relation edge, and storing the key map after integral encryption.
  7. 7. The method for constructing and encrypting a large model-based distributed knowledge graph according to claim 6, wherein the step S5 further comprises a step S6 of secure query processing: receiving a query request, analyzing the query intention to determine an entity and a relation set which need to be accessed; Positioning a corresponding encryption key according to the key map, and acquiring a corresponding encryption data block from the distributed storage node; and (5) performing decryption and graph traversal calculation in a secure environment, and returning a query result.
  8. 8. The method for constructing and encrypting a large model-based distributed knowledge graph according to claim 1, wherein the knowledge fusion in step S4 includes conflict detection and resolution of attributes from different large model agents directed to the same entity.
  9. 9. The large-model-based distributed knowledge graph construction and encryption storage method according to claim 1, wherein the method is characterized in that encryption transmission is performed when the preliminary knowledge segments generated in step S2 are transmitted between distributed nodes.
  10. 10. The large model-based distributed knowledge graph construction and encryption storage method according to claim 1, further comprising the step of S7, incremental update and key rotation: When new knowledge is needed to be added to the encrypted stored knowledge graph in an increment mode, an encryption key is generated for the new knowledge, and meanwhile, the encryption key of the related original knowledge affected is triggered to be updated in a rotation mode.

Description

Distributed knowledge graph construction and encryption storage method based on large model Technical Field The invention relates to the technical field of knowledge maps, in particular to a distributed knowledge map construction and encryption storage method based on a large model. Background The knowledge graph is used as a semantic network for describing the relation between entities and is a basic stone for key application of artificial intelligence. The traditional construction method relies on expert manual or rules, and has low automation degree and poor expansibility. Although the breakthrough of large language models and other technologies brings hopes for automatic construction, knowledge can be extracted from massive texts, two major challenges are faced, namely firstly, the inherent 'illusion' problem of the large models can cause inaccurate or fictional knowledge to be generated and seriously damage the credibility of the atlas, secondly, an effective collaboration mechanism is lacking at present, a plurality of large model examples cannot be comprehensively constructed in a distributed and verifiable mode, and consistency of a knowledge acquisition process and quality of final results are difficult to ensure. These factors limit the mass production of high quality knowledge patterns. In terms of storage, the distributed system can solve the problem of expansibility of massive knowledge data, but also introduces serious potential safety hazards. The map often contains sensitive information, and the plaintext is stored in the distributed node and is easy to be accessed and leaked by unauthorized. The existing encryption scheme is often disjointed with the construction flow and mostly remedied afterwards, and faces the dilemma that coarse-granularity encryption seriously damages the query efficiency, fine-granularity encryption is complex in key management and difficult to support efficient multi-hop query. Therefore, a scheme capable of integrating high-quality construction with secure storage is needed in the prior art to solve the problem of full-link from distributed collaborative construction, quality control to encrypted storage. Disclosure of Invention The invention particularly relates to a large-model-based distributed knowledge graph construction and encryption storage method, which aims to solve the problems of uncontrollable knowledge graph construction quality, low automation degree and insufficient distributed storage safety in the prior art. In order to achieve the above purpose, the specific technical scheme of the distributed knowledge graph construction and encryption storage method based on the large model is as follows: a distributed knowledge graph construction and encryption storage method based on a large model comprises the following steps: Step S1, task planning and agent allocation, namely receiving a knowledge graph construction request, decomposing an overall construction task into a plurality of knowledge extraction subtasks based on a predefined ontology mode, and allocating the subtasks to a plurality of large model agents with corresponding high-priority energy in a distributed network according to a pre-constructed large model agent capability image; s2, carrying out entity identification, relation extraction and attribute filling operations from a designated unstructured or semi-structured data source by each large-model intelligent agent to generate a structured preliminary knowledge segment; step S3, multiple rounds of cross-validation and consensus are achieved, namely, multiple rounds of cross-validation are initiated in a distributed intelligent agent network aiming at the preliminary knowledge segments, the knowledge segments with association or conflict are distributed to at least one other large model intelligent agent for validation, and the knowledge segments with high credibility are screened out based on the confidence and consistency of the validation results to achieve consensus; S4, carrying out logic fusion on the high-credibility knowledge segments which reach consensus, eliminating conflict and forming an internally consistent high-quality quantum spectrum; And S5, carrying out fine-granularity dynamic encryption storage, namely carrying out security level assessment on each entity and relation in the sub-map, dynamically generating an encryption key based on the assessment result and the structural context of the encryption key in the sub-map, carrying out fine-granularity encryption, and finally storing the encrypted data blocks in a plurality of storage nodes in a distributed mode. Further, decomposing the overall construction task based on the predefined ontology schema in the step S1 specifically includes vertically dividing the construction task according to entity types and relationship types defined in the ontology schema, and/or horizontally dividing the construction task according to data sources to form the knowledge extraction subtas