Cluster Labeling for Patent Panoramic Analysis

Abstract

Patent data is important for companies to grasp the technology trends and the status of competitors. Patent panoramic analysis supports companies to position themselves in the market and grasp the trends of competitors, because it enables users to overview patents in a specified technical field by clustering and visualizing them. In each cluster, the users can capture the technical features of the patents by showing them cluster labels (representative keywords in the cluster). However, cluster labels often overlap among the clusters when we try to assign labels which consists of high frequency words in each cluster. This is because words that appear frequently in one cluster also appear frequently in other clusters when we cluster patents in a specified technical field. To tackle this challenge, we propose to use phrases as cluster labels and extract optimal combinations of the phrases which are discriminative among the clusters. In the experiment, we evaluate the effectiveness of our proposed method by using patent datasets.