Measuring similarity between patents is an essential step to ensure novelty of innovation. However, a large number of methods of measuring the similarity between patents still rely on manual classification of patents by experts. Another body of research has proposed automated methods; nevertheless, most of it solely focuses on the semantic similarity of patents. In order to tackle these limitations, we propose a hybrid method for automatically measuring the similarity between patents, considering both semantic and technological similarities. We measure the semantic similarity based on patent texts using BERT, calculate the technological similarity with IPC codes using Jaccard similarity, and perform hybridization by assigning weights to the two similarity methods. Our evaluation result demonstrates that the proposed method outperforms the baseline that considers the semantic similarity only.

翻译：测量专利间的相似性是确保创新新颖性的关键步骤。然而，大量专利相似度测量方法仍依赖专家人工分类。另一类研究提出了自动化方法，但其中多数仅关注专利的语义相似度。为解决这些局限，我们提出了一种混合方法，可同时兼顾语义相似性与技术相似性，实现专利相似度的自动测量。该方法基于专利文本使用BERT测量语义相似度，利用IPC代码通过Jaccard相似度计算技术相似度，并通过为两种相似度方法分配权重进行混合。评估结果表明，所提方法优于仅考虑语义相似度的基线方法。