HKUDS/RLMRec

Leverages LLM-generated user/item profiles and semantic embeddings to bridge collaborative filtering with semantic understanding, using contrastive or generative alignment to synchronize LLM representation spaces with graph-based collaborative signals. Implements a model-agnostic framework compatible with multiple graph neural network backbones (LightGCN, SGL, SimGCL, etc.) on text-attributed recommendation datasets. Built on PyTorch with pre-processed datasets (Amazon-book, Yelp, Steam) that include LLM-generated profiles, text embeddings, and sparse interaction matrices for reproducibility.

448 stars. No commits in the last 6 months.