Technology analysis from patent data using latent dirichlet allocation

This paper discusses how to apply latent Dirichlet allocation, a topic model, in a trend analysis methodology that exploits patent information. To accomplish this, text mining is used to convert unstructured patent documents into structured data. Next, the term frequency-inverse document frequency (tf-idf) value is used in the feature selection process. After the text preprocessing, the number of topics is decided using the perplexity value. In this study, we employed U.S. patent data on technology that reduces greenhouse gases. We extracted words from 50 relevant topics and showed that these topics are highly meaningful in explaining trends per period.