| Project Reference: | ITP/001/24LP |
| Project Title: | Local-Trained Large Language Model with Domain- Specific Scalable Concept Co-occurrence Relationship Models for Drafting Comprehensive and Balanced Public Articles and Messages |
| Hosting Institution: | LSCM R&D Centre (LSCM) |
| Abstract: | The proposed R&D work is to develop a local-trained large language model (LLM) adherence to the information sources at Health Bureau and health-related authoritative health organizations like World Health Organization and US Centers for Disease Control and Prevention. The trained LLM can compose various drafts according to corresponding user queries. Together with the composing capability of the LLM, the traceability of information sources is validated by an information retrieval mechanism supported by a set of graph-based concept co-occurrence relationship models (CCRMs) being developed to have traceable links among terms, co-occurred term pairs, and source information items. A set of relevant source information items retrieved per user query from CCRMs helps establish an information scope to validate corresponding generated draft from the LLM. In addition, all drafts with different aspects can be compared with corresponding traceable information sources for further refining to yield a well-balanced message being comprehensive enough to disseminate to the public suitable for the context at that moment. Thus, the proposed R&D is expected to all Health Bureau officials to obtain a set of source-supported drafts from their queries and then spend time on crafting comprehensive and balanced public articles and messages. |
| Project Coordinator: | Dr To Bun Ng |
| Approved Funding Amount: | HK$ 14.29M |
| Project Period: | 28 Mar 2024 - 27 Mar 2026 |
