Research of the LDA algorithm processing results on high- level classes of patents

Abstract

The purpose of the article is to study the similarity of extractable topics from different high-level classes of patents and the possibility of classifying these documents according to the generally-trained model. The optimal number of topics can be selected from the interpretation of the resulting topics for the coherence of words in the topic and the reflection of the general discourse. In the presented dataset only general themes are known, is not possible to suggest which sub-themes can discover. In the course of the research, the dynamics of the change in the models quality with the change of parameters, according to which relatively optimal parameters are chosen, is considered.