Reproduce “MemWeaver: A Hierarchical Memory from Textual Interactive Behaviors for Personalized Generation” 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
Shuo Yu, Mingyue Cheng, Daoyu Wang, Qi Liu, Zirui Liu, Ze Guo, Xiaoyu Tao
Researchers at the University of Science and Technology of China developed MemWeaver, a cognition-inspired framework that constructs a hierarchical memory from user textual interactions to guide personalized content generation by Large Language Models. This system achieved state-of-the-art performance on the LaMP benchmark, outperforming baselines across all six datasets and twelve evaluation metrics.
Key insights
- Deep personalization requires moving beyond flat user histories to models that capture the rich temporal evolution and intricate semantic relationships within textual interactive behaviors.
- A cognition-inspired hierarchical memory structure, distinguishing between concrete behaviors and abstract cognitive traits, offers a more comprehensive user understanding than conventional methods.
- Jointly leveraging dual memory components (behavioral and cognitive) enables Large Language Models to balance immediate context with long-term preferences for effective personalized generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.