Reproduce “A Universal Sets-level Optimization Framework for Next Set 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
Yuli Liu, Min Liu, Christian Walder, Lexing Xie
SNSRec predicts next-set recommendations by modeling temporal sets as units rather than ranking items alone. It uses a structured determinantal point process to combine item relevance, within-set co-occurrence, and category-aware diversity, training on sampled negative sets. On JD-add, it scores 0.2192 F1@20 versus 0.1845 for DSNTSP, a reported 18.78% improvement in offline benchmarks, not evidence of real-world user outcomes.
Key insights
- Optimizing directly at the set entity level, rather than using item-level objectives, more effectively captures complex dependencies within and across sequential sets for next set recommendation.
- Explicitly incorporating both co-occurrence patterns (intra-set cohesion) and diversity metrics into the optimization criterion improves both relevance and variety of recommended sets.
- The proposed framework is universal, demonstrated by its ability to integrate with and significantly enhance the performance of diverse existing Next Set Recommendation architectures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.