Count of processed information has incredibly increased last time. This phenomenon has made necessity in thematic modeling systems, which could help that information to systematize. This paper presents the news classifier which is based on the Online LDA algorithm. The main peculiarity of the classifier is the assignment of the several most relevant categories of each text. Learning and tests was doing on Reuters news collection, which include 21578 documents. Despite some modest results, this method might be useful in practice because operating speed in the education and runtime modes is sufficiently high. In addition, performance could be raised substantially by making more accurate selection of hyperparameters.