TOPIC MODELING IN TEXT BASED ON TITLES OF DOCUMENTS

Abstract

The paper presents an approach for topic modeling and document classification. Concretely, the paper explores 1) the application of LDA (Latent Dirichlet Allocation) to obtain topics from the text; This approach was evaluated qualitatively, relying on the semantics of the found topics. 2) document classification, where documents were represented using tf-idf features and topics extracted by applying LSA (Latent Semantic Analysis); Naive Bayes classifier was trained on the obtained representation, and F-measure was used for evaluation, 3) document classification, where the tf-idf representation was used to train a classification model, where we experimented with using SVM (Support Vector Machines) and RF (Random Forest) models; In this case, F-measure was used for evaluation.