Technology Forecasting using Topic-Based Patent Analysis

Abstract

The number of patents with critical information related to various technologies is increasing by the day. This trend has led corporations and countries to consider patent analysis as an important element in their analysis methodology for research and development. The present study seeks to determine and forecast vacant technology with considerable development potential through an analysis of patents. In order to identify a vacant technology cluster, the unstructured patent documents need to be structured into groups of similar technologies by using k-means clustering. Furthermore, silhouette width, Davies-Bouldin Index (DBI), and Pseudo F are used for enhancing reliability of determining the optimal number of clusters. From each technology cluster, a generative topic model, latent Dirichlet allocation (LDA), is adopted to extract latent topics specifically for examination of technologies. Renewable energy patents from the United States Patent and Trademark Office (USPTO) are analyzed for the case study, which verifies the proposed methodology.