Reproduce “Behavioral Feature Boosting via Substitute Relationships for E-commerce Search” 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
Chaosheng Dong, Michinari Momma, Yijia Wang, Yan Gao, Yi Sun
Amazon researchers developed a Behavior Feature Boosting via Substitute Relationships (BFS) method to address the cold-start problem for new products in e-commerce search, enhancing their discoverability. The approach imputes behavioral signals like Sales Velocity from established substitute products, resulting in statistically significant online A/B test gains, including +0.11% in Gross Merchandise Volume and +0.35% in New Product Sales Units.
Key insights
- Leveraging substitute relationships, identified via multimodal embeddings and classification, effectively provides proxy behavioral signals for new products lacking direct interaction data.
- Aggregating Sales Velocity from established substitutes significantly boosts the perceived relevance of new items in search rankings.
- The boosted behavioral feature (SV_Subs) increases the search relevance score for under-discovered products, especially those below a certain Sales Velocity threshold, without degrading performance for existing popular items.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.