LDA models for finding trends in technical knowledge domain

The search of latent dependencies and the analysis of text collections are important tasks in the context of rapid development of information technologies. This article describes the identification of scientific and technological trends in the field of it-technologies using the analysis of thematic publications from web sources with LDA topic models. The proposed solution uses a formal criterion to identify the best results among the topic models. This criterion is based on the competitive analysis of search queries and automatically highlights trends. Expert evaluation of the LDA models and automated verification of these evaluations were made. Computational experiment shows that the expert evaluation coincides with results of formal criterion in general. This method is suitable for finding trends in technical knowledge domain.