Reproduce “Knowledge-aware Complementary Product Representation Learning” 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
Da Xu, Chuanwei Ruan, Jason Cho, Evren Korpeoglu, Sushant Kumar, Kannan Achan
Knowledge-aware Complementary Product Representation Learning separates personalized preference from directional product complementarity, using purchase sequences and product context to rank items that work well together. Context provides the clearest gains on sparse Walmart data, where the model reaches Hit@10 of 0.2518 and NDCG@10 of 0.2139, and it can also infer representations for items with no purchase history.
Key insights
- Modeling preference separately from complementarity helps distinguish what a customer generally likes from what goes well with their recent purchases.
- Product context is particularly useful when interaction data are sparse: removing item context substantially reduced recommendation performance on the large Walmart dataset.
- Context does not improve every metric in every setting: on Instacart, the full model had higher AUC than its no-context ablation, but lower NDCG.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.