News Topic Categorization on Telugu Dataset Using Deep Learning Models

The digital age has ushered in an unprecedented era of data proliferation across various domains, with the news sector being no exception, accumulating vast amounts of information, often measured in petabytes. Manual classification of such voluminous data is impractical, particularly considering the complexity of languages like Telugu, belonging to the Dravidian family, renowned for its intricate morphological characteristics. Natural Language Processing (NLP) offers a promising avenue for tackling textual data, leveraging diverse word embedding techniques to enrich model representations. Notably, prior research in this domain has predominantly relied on count-based algorithms for word embeddings. However, in this investigation, we explore alternative approaches, specifically custom-built Deep Learning models integrated with word embeddings, to optimize the classification of newspaper articles. Through a series of methodical experiments, we endeavor to identify the most effective model architecture for this task, aiming to enhance the accuracy and efficiency of article categorization in Telugu. The proposed model has experimented with word embeddings such as FastText and GloVE with deep learning models like LSTM, Bi-LSTM, CNN. Among the models experimented, FastText in combination with CNN have reported the highest F1-Score of 0.93.