A Scientometric Study of Technological Trend Based on Patent

Abstract

Different from traditional statistical analysis models which focus on the correlation between technical trends in quantity level, text analyzing is now a new approach to extract technological trend from text. Scientometric text mining is widely applied in analytical methods to figure the evolution process of patent and technology. This study is focused on patent documents to predict the future trend of solid material field. Term Frequency (TF) statistics, Word2Vec and t-SNE were adopted in comprehensive analytical methods to measure metrics of patent and reveal the technological development. For patent documents in No. 257 category of United States Patent and Trademark Office, title, abstract and claims from 2005 to 2012 were selected for text mining and analysis. The term frequency of those keywords is firstly counted by year, to extract the annual change of high-frequency keywords. Word2Vec can convert the text to word vectors in a vector space. To better visualize the results and make a relatively reliable prediction, t-SNE is used to reduce the dimension word vectors and scatter them in a twodimensional map. All these methods in the research are dedicated to present a clearer look at the evolving trend of patents in a field. The systematic analytical methods can be adapted in the analysis of other fields. What is represented in the results reveal the developing trend and the hotspots in process, thereby the future trend can be inferred based on the results. Such prediction can be an indicator or guidance in decision making of a certain field.