Automatic Labeling of Topic Models

Topic modeling is a technology of statistical text analysis that can automatically extract topics from large collections of documents. In order to interpret topics after the modeling and represent them in information retrieval and recommendation services, It is necessary to solve the problem of automatic labeling of topics. This paper proposes a new approach to labeling top-level topics in a hierarchical topic model. The approach is based on using an additional hierarchical model built on the Wikipedia using two modalities: texts of the articles and their category lists. An optimization algorithm is proposed for matching topics of two hierarchical models. It takes into account not only the proximity between pairs of topics, but also their structural relations in the model hierarchy. Once matching is complete, it is proposed to choose the top categories of Wikipedia topics as label candidates for the unlabelled topic model. This method was tested on news articles, and, as a result, suitable labels for the top-level topics were generated.