Researchers have introduced MemHop, a new benchmark designed to evaluate the multi-hop reasoning capabilities of Large Language Model (LLM) agents. This benchmark consists of 1,000 questions with evidence annotations across 10 social network scenarios, testing recall at depths of 1 to 5 hops. Alongside MemHop, the paper proposes Profile-Graph Memory (ProGraph), a novel memory architecture that combines profile expansion for traversing entity names within narrative profiles and compression residuals for capturing precise details like dates and quantities. ProGraph demonstrated strong performance on both MemHop and the Long Context Modeling (LoCoMo) benchmark, outperforming existing memory frameworks. AI
IMPACT Introduces new methods for evaluating and improving LLM agent long-term memory, potentially enhancing their ability to handle complex, multi-step tasks.
RANK_REASON Academic paper introducing a new benchmark and memory architecture for LLM agents. [lever_c_demoted from research: ic=1 ai=1.0]
- A Memoir Blue
- arXiv
- FullContext
- HippoRAG
- LLM agents
- Long Context Modeling
- Mem0 Agent Memory Framework
- MemHop
- Profile-Graph Memory
- ProGraph
- retrieval-augmented generation
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
