Conference

Co-occurrence based predictors for estimating query difficulty

This paper explores query difficulty prediction to assess how reliably an information retrieval (IR) system can handle user queries. Word mismatches between user queries and document vocabulary often hinder retrieval effectiveness, and while query expansion (QE) helps, its impact varies across different query sets. To address this inconsistency, the study investigates query difficulty predictors, introducing two new predictors based on the co-occurrence of query terms. Experiments on TREC collections evaluate their effectiveness, contributing to enhancing query processing reliability and ensuring more robust query expansion strategies.