Word Embeddings Versus LDA for Topic Assignment in Documents

Abstract

Topic assignment for a corpus of documents is a task of natural language processing (NLP). One of the noted and well studied methods is Latent Dirichlet Allocation (LDA) where statistical methods are applied. On the other hand applying deep-learning paradigm proved useful for many NLP tasks such as classification [3], sentiment analysis [8], text summarization [11]. This paper compares the results of LDA method and application of representations provided by Word2Vec [5] which makes use of deep learning paradigm.