abdoelsayed/FutureQueryEval

FutureQueryEval Dataset (EMNLP 2025)🔍 Dataset Description FutureQueryEval is a novel Information Retrieval (IR) benchmark designed to evaluate reranker performance on temporal novelty. It comprises 148 queries with 2,938 query-document pairs across 7 topical categories, specifically created to test how well reranking models generalize to truly novel queries that were unseen during LLM pretraining. Key Features Zero Contamination: All queries refer… See the full description on the dataset page: https://huggingface.co/datasets/abdoelsayed/FutureQueryEval.