Exploring Topic Modelling Techniques on Indian News: a Comparitive Study

This study aims to understand the unique strengths and limitations of the four topic modelling approaches: Latent Dirichlet Allocation (LDA), Hierarchical Dirichlet Process (HDP), Non-Negative Matrix Factorization (NMF) and Latent Semantic Analysis (LSA). These approaches will be applied on extracted News datasets from two news outlets which are among some of the most prominent in India: News18 and NDTV. The focus is more specified onto World Category. By comparing these methods in our research, we evaluate their effectiveness in uncovering various thematic structures, trends and large-scale textual data is interpreted. As a result, both quantitative and qualitative metrics are used to assess the performance, keeping in mind the emphasis on scalability and interpretability. Learning how these algorithms function helps us to understand how latent topics are being extracted from a large amount of textual data. As a result, each of the thematic structure brought into light gives a new understanding on topic modelling and their computational complexities in handling different data from different news outlets.