MedGraph
A graph-memory-based memory system for LLM medical dialogue. By constructing fact chains with timestamps and person attribution, it reads states at the time point of a question during retrieval, addressing state tracking, person attribution, and safe fact recall in long-term medical conversations.
Timeline (last 90 days)
MedGraph builds graph memory that links facts into chains by date and person attribution, and reads them at the time point of the question during retrieval.
On the MedMemoryBench benchmark, MedGraph with a GPT backbone achieves the highest overall accuracy, leading the strongest baseline by 10.5 percentage points.
On the MedMemoryBench benchmark, MedGraph with a Qwen backbone achieves the highest overall accuracy, leading the strongest baseline by 10.4 percentage points.
Under the GPT setting, MedGraph's memory construction and answering efficiency are also relatively high.
MedGraph's patient profile lists all recorded safety facts and associates them with each question.
All Facts (5)
MedGraph builds graph memory that links facts into chains by date and person attribution, and reads them at the time point of the question during retrieval.
50%UnverifiedOn the MedMemoryBench benchmark, MedGraph with a GPT backbone achieves the highest overall accuracy, leading the strongest baseline by 10.5 percentage points.
50%UnverifiedOn the MedMemoryBench benchmark, MedGraph with a Qwen backbone achieves the highest overall accuracy, leading the strongest baseline by 10.4 percentage points.
50%UnverifiedUnder the GPT setting, MedGraph's memory construction and answering efficiency are also relatively high.
50%UnverifiedMedGraph's patient profile lists all recorded safety facts and associates them with each question.
50%