Comparing Patent Network Approaches for Extracting Technology Intelligence
比较了基于引用、分类和文本的三种专利相似度测量方法在网络结构、聚类和关键专利识别上的差异，为选择构建专利网络的策略提供指导。
Patent network offers advantages for extracting technology intelligence, such as analyzing technology trend and identifying emerging technologies. Despite the significant value of patent network methods, few studies compare similarity measures used to construct a network, which can have a substantial impact on analysis results. Because technology intelligence extraction depends on contextual relevance, understanding the unique characteristics of each similarity measure enables more appropriate and effective insights. This study first conducts a network structure analysis using multiple patent similarity measures. Basic network statistics and comparisons of network similarity are performed, along with sensitivity analyses that vary similarity measurement levels and threshold settings to examine how network characteristics change under different analytical conditions. The study then compared three similarity measures—citation-, classification-, and text-based approaches—across two perspectives: patent clustering and key patent identification. For clustering, we further evaluate robustness, assessing stability across hyperparameter configurations and consistency of pairwise assignments. The findings provide guidance on the selection of patent network construction strategies for analytical objectives.