DigiNews

Summary

The article introduces Bag of Decisions Reranker, a method that uses an LLM to generate a rubric of relevance questions and combines it with Jev reranking to evaluate per-query relevance. It presents experimental results showing improved mean NDCG over BM25 and Jev reranker on Wayfair WANDS and Amazon ESCI datasets, and discusses potential domain-focused extensions. It also promotes an AI-powered search course.