Using Machine Learning Approaches to Identify Emergence: Case of Vehicle Related Patent Data

Bibliometric studies have long used simple search strings, publications count, and word counts to track the emergence of technologies. Novel machine learning methods open new possibilities to study bibliometric data and use algorithmic approaches to uncover emergence of a technology. This study looks at the large and complex dataset of vehicle related patents to uncover emergence indicators. By using machine learning methods this study focuses on if, and to what extent different methods can produce patterns of emergence from the data directly. The data extracted from PATSTAT contains 711296 granted US patent abstracts between the years 1980 and 2014 resulting from a search for "vehicle" creating a complex dataset of technologies from automotive to medical applications. Using Latent Dirichlet Allocation and Dynamic Topic Modeling we show different emergence patterns. Finally, we discuss in detail the possibilities of using machine learning approaches to draw emergence dynamics of technologies.