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

Figure CN122432980A_ABST