A multi-source heterogeneous data knowledge extraction method for a large language model

By establishing heterogeneous networks and heterogeneous knowledge graphs for multimodal data, the problem of insufficient multimodal data modeling capabilities was solved, enabling effective knowledge extraction and analysis of multi-source heterogeneous data and improving the semantic information expression capabilities of large language models.

CN122432980APending Publication Date: 2026-07-21NAVAL UNIV OF ENG PLA
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
NAVAL UNIV OF ENG PLA
Filing Date
2026-04-23
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Traditional data-focusing methods struggle to achieve unified knowledge data modeling across multimodal data, limiting their ability to express semantic information and intrinsic connections.

Method used

By establishing heterogeneous networks and utilizing the heterogeneous network paths formed by data and data relationships, as well as the inertia between points, knowledge extraction is performed on multi-source heterogeneous data. This includes data collection, preprocessing, cross-modal alignment, multimodal knowledge attribute feature representation, and the establishment of heterogeneous knowledge graphs for multimodal data.

Benefits of technology

It enables effective knowledge processing and analysis of multi-source heterogeneous data in multimodal large language models, and improves the semantic information expression capability of multimodal data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122432980A_ABST
    Figure CN122432980A_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of large language model data processing, and particularly relates to a multi-source heterogeneous data knowledge extraction method for a large language model. The application comprises data collection, data preprocessing, multi-modal knowledge attribute feature expression, definition of core nodes and attribute nodes based on a multi-modal heterogeneous knowledge graph, configuration of a heterogeneous knowledge graph structure path, distribution of path weights, establishment of a multi-modal heterogeneous knowledge graph, calculation of a score to obtain path fusion weights, and further obtaining of knowledge expression of a target node. The application considers multi-modal data joint expression, establishes a heterogeneous network, uses a heterogeneous network path formed by data and data relationships and inertia between points to further mine internal knowledge between heterogeneous data, and provides a knowledge data processing and analysis scheme for establishment of a multi-modal large language model.
Need to check novelty before this filing date? Find Prior Art