Reproduce “From Retrieval to Reconstruction: Constructing Evolvable Cognitive Memory for Long-Term Dialogue” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zirui Liao, Zhengxian Wu, Zhuohong Chen, Yunyao Yu, Xiaoyu Liu, Yifan Xu, Haoqian Wang
CogMem helps long-term dialogue assistants recover events and attributed opinions by storing dialogue in a typed graph and actively reconstructing answers. Its records distinguish Events, Facts, and source-attributed Claims, while retrieval selects graph operators and returns source evidence. On LoCoMo Multi-Hop with Qwen2.5-14B, it scores 48.42% word-level-overlap F1 versus GAM’s 42.96%, within two English benchmarks and two backbones.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.