Reproduce “Context-Aware Sequential Model for Multi-Behaviour Recommendation” 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
Shereen Elsayed, Ahmed Rashed, Lars Schmidt-Thieme
CASM is a compact sequential recommender that augments chronological item histories with behavior-type context and behavior-specific training weights. Across Taobao, Tianchi, Yelp, and MovieLens, it achieved the strongest reported HR@10 and NDCG@10 results, while remaining substantially faster than several graph-based multi-behavior models.
Key insights
- Auxiliary interactions are particularly valuable when users have few target actions; on Tianchi, incorporating them raised HR from about 0.245 to 0.310 and NDCG from 0.160 to 0.230 for sparse-purchase users.
- Manually chosen behavior weights matter more on e-commerce data, where views, favorites, carts, and purchases represent different intent levels; dislike signals contributed little on MovieLens.
- Temporal self-attention and multi-behavior modeling complement one another, but the best reported configuration used only one attention block and one head, emphasizing efficiency over architectural scale.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.