PatSnap/novelty-search-bench
收藏资源简介:
PatSnap新颖性搜索基准是一个用于评估专利新颖性搜索(现有技术搜索)系统的基准数据集。该数据集包含569个样本,每个样本包括一个查询专利的公开号(PN)以及由专利审查员识别的X型(破坏新颖性)现有技术引用作为真实标签。数据集设计为内部完整评估集的50%公开版本,结合了两种互补的样本类型:跨管辖区域家族扩展样本和单管辖区域非扩展样本,以便在标准非扩展基线之外,还能对信息量最大的现有技术场景进行检索系统测试。数据来源于真实的专利审查记录,覆盖CN、US、EP、WO、AU等多个管辖区域,语言包括英语(55.0%)和中文(45.0%),IPC分类覆盖所有8个部分(A-H)。每个样本提供了查询专利的完整描述文本,可直接用于检索系统输入,无需外部查找。数据集还提供了详细的字段说明、真实标签构建方法、评估指标(如top@K和Recall@K)以及样本分布统计(按类型、管辖区域、语言和IPC分类)。
PatSnap Novelty Search Bench is a benchmark for evaluating patent novelty search (prior art search) systems. Each sample contains a query patent publication number (PN) along with ground truth X-type (novelty-destroying) prior art references identified by patent examiners. The Bench is deliberately designed as a 50% public release of an internal full evaluation set that combines two complementary sample types — cross-jurisdiction family-expanded samples and single-jurisdiction non-expanded samples — so that retrieval systems can be probed on the most informative prior-art scenarios alongside a standard non-expanded baseline. The data is sourced from real patent examination records, covering jurisdictions such as CN, US, EP, WO, and AU, with languages including English (55.0%) and Chinese (45.0%), and IPC coverage across all 8 sections (A–H). Each sample includes the full description text of the query patent, ready to be fed directly into a retrieval system without external lookup. The dataset provides detailed field descriptions, ground truth construction methods, evaluation metrics (e.g., top@K and Recall@K), and distribution statistics (by sample type, jurisdiction, language, and IPC section).




