THEORETICAL FOUNDATIONS AND CONCEPTS OF TOPIC MODELING

Topic modeling is a powerful tool for automatically identifying hidden topics in collections of text documents. This paper reviews the main topic modeling methods, ranging from classical approaches such as Latent Semantic Analysis (LSA) and NonNegative Matrix Factorization (NNMF) to stateoftheart neural network methods including Neural Topic Modeling (NTM) and BERTopic. Particular attention is paid to the evolution of the methods, from probabilistic models (PLSA, LDA) to flexible regularized approaches (ARTM) and neural network architectures based on variational autoencoders. The advantages and limitations of each method are analyzed, as well as their application in natural language processing tasks. The article is intended for researchers and practitioners interested in modern methods of text data analysis.