PatentMatch

我们通过创建一个名为 PatentMatch 的监督机器学习训练数据集来解决现有技术的计算机辅助搜索问题。它包含来自专利申请的成对的权利要求和来自引用的专利文件的不同程度的语义对应文本段落。每一对都由欧洲专利局技术娴熟的专利审查员贴上标签。因此，标签表示语义对应（匹配）的程度，即文本段落是否有损于要求保护的发明的新颖性。We address the computer-aided search problem in the prior art by constructing a supervised machine learning training dataset named PatentMatch. This dataset comprises paired text segments: one being the claims from patent applications and the other being text paragraphs from cited patent documents, with varying degrees of semantic correspondence between each pair. Each pair is labeled by highly skilled patent examiners from the European Patent Office (EPO). Accordingly, the labels indicate the degree of semantic correspondence (i.e., matching), namely whether the text paragraphs would undermine the novelty of the claimed invention.

搜集汇总

背景与挑战

背景概述

PatentMatch是一个用于计算机辅助专利搜索的监督学习数据集，包含专利申请权利要求与引用专利文本段落的匹配对，由专利审查员标注匹配程度。该数据集由波茨坦大学于2020年发布，适用于文本分类和挖掘任务。