Drawing on cognitive psychology, the authors define three memory types for LLM agents: episodic (events with time/place/participants), personal semantic (stable facts about the user), and general semantic (common knowledge).
Classifying memory by type improves LLM retrieval for long-term conversations
MemoType dynamically routes memories and queries to tailored retrieval strategies, outperforming unified approaches by up to 16% in recall.
Chinese Tech
Yi Wen · Derong Xu · Pengyue Jia · Yichao Wang · Yingyi Zhang · Maolin Wang · +5 more
City University of Hong Kong · Huawei Noah’s Ark Lab
Research Digest··2 min read
The authors introduce MemoType, a framework that classifies conversational memories into three types—episodic, personal semantic, and general semantic—and applies retrieval strategies customized to each type.
Why this paper
From Huawei Noah’s Ark Lab and City University of Hong Kong
In one line
Tailored retrieval strategies per memory type outperform unified approaches in LLM agent long-term memory.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§