Formbench-anon/FormBench

FormBench是一个大规模的信息检索基准，专注于配方科学，包括粘合剂、涂料、聚合物和制药等多个行业。数据集包含约100万条语料库段落、55,352条查询和4级分级相关性qrels，这些qrels源自590K美国配方专利的领域分类法。数据集提供三种配置：formbench-structured、formbench-random和formbench-sample，每种配置采用不同的段落选择策略。数据集还包括分级相关性评分、文件模式、领域分类法、负责任的AI考虑因素和维护计划的详细信息。

FormBench is a large-scale information retrieval benchmark for formulation science, covering industries such as adhesives, coatings, polymers, and pharmaceuticals. It provides approximately 1M corpus passages, 55,352 queries, and 4-level graded relevance qrels derived from a domain taxonomy of 590K US formulation patents. The dataset is available in three configurations: formbench-structured, formbench-random, and formbench-sample, each with different passage selection strategies. It also includes detailed information on graded relevance scores, file schema, domain taxonomy, responsible AI considerations, and maintenance plans.