HyperMem: Hypergraph Memory for Long-Term Conversations

2026-04-09

AI Summary

Existing dialogue systems struggle to model high-order interactions among multiple elements, leading to fragmented memory representations that undermine conversational coherence and personalization. To address this limitation, this work proposes a hypergraph-based hierarchical memory architecture that, for the first time, introduces hypergraphs into dialogue memory modeling to explicitly capture joint dependencies among multiple entities. The memory is structured into three levels—topics, events, and facts—and coherent memory units are formed through hyperedge aggregation. Coupled with a hybrid lexical-semantic indexing scheme and a coarse-to-fine retrieval strategy, the approach enables efficient and accurate recall of high-order associations. Evaluated on the LoCoMo benchmark, the method achieves a state-of-the-art performance with a 92.73% accuracy under LLM-as-a-judge evaluation, significantly outperforming existing approaches.

Abstract

Long-term memory is essential for conversational agents to maintain coherence, track persistent tasks, and provide personalized interactions across extended dialogues. However, existing approaches as Retrieval-Augmented Generation (RAG) and graph-based memory mostly rely on pairwise relations, which can hardly capture high-order associations, i.e., joint dependencies among multiple elements, causing fragmented retrieval. To this end, we propose HyperMem, a hypergraph-based hierarchical memory architecture that explicitly models such associations using hyperedges. Particularly, HyperMem structures memory into three levels: topics, episodes, and facts, and groups related episodes and their facts via hyperedges, unifying scattered content into coherent units. Leveraging this structure, we design a hybrid lexical-semantic index and a coarse-to-fine retrieval strategy, supporting accurate and efficient retrieval of high-order associations. Experiments on the LoCoMo benchmark show that HyperMem achieves state-of-the-art performance with 92.73% LLM-as-a-judge accuracy, demonstrating the effectiveness of HyperMem for long-term conversations.

Problem

• long-term memory
• conversational agents
• high-order associations
• fragmented retrieval
• hypergraph

Innovation

• hypergraph memory
• high-order associations
• hierarchical memory architecture
• coarse-to-fine retrieval
• long-term conversation