ZooWork-ShopRanker: An Open, Preference-Aligned E-Commerce Reranker
ZooWork-ShopRanker uses LLM-judged shopping preferences to train open e-commerce rerankers in 0.6B, 4B, and 8B sizes.
The paper says general web rerankers do not transfer cleanly to shopping, where product fit and constraints matter alongside relevance. Its training labels come from multiple reasoning LLM families acting as a preference oracle, with debiasing and agreement tiers. The 8B model is then used to distill smaller 4B and 0.6B versions. The authors also introduce ShopRank-Bench, about 10,000 private-traffic preference pairs, and report gains over open baselines and unaligned bases. HF Daily Papers' note
The paper says general web rerankers do not transfer cleanly to shopping, where product fit and constraints matter alongside relevance. Its training labels come from multiple reasoning LLM families acting as a preference oracle, with debiasing and agreement tiers. The 8B model is then used to distill smaller 4B and 0.6B versions. The authors also introduce ShopRank-Bench, about 10,000 private-traffic preference pairs, and report gains over open baselines and unaligned bases. HF Daily Papers' note
score 4