Exploring Deep Learning Approaches for News Classification with CNNs, RNNs and Transformers

In the realm of text classification, particularly news classification, advanced deep learning models have demonstrated substantial potential for enhancing accuracy and performance. This paper investigates the application of several prominent deep learning architectures: CNNs, RNNs and Transformer-based BERT model. CNNs were employed to capture local patterns and hierarchical features within text data, contributing to effective categorization of news articles. RNNs, including LSTM networks, were utilized to model sequential dependencies and contextual relationships in news text, addressing the challenge of understanding temporal aspects. Additionally, Transformer-based models such as BERT (Bidirectional Encoder Representations from Transformers) were leveraged for their superior contextual and semantic comprehension, achieving notable performance improvements. The study provides a comprehensive analysis of these models' effectiveness, comparing their strengths and limitations based on accuracy results. BERT and its hybrid combinations, particularly with LSTM, achieved the highest accuracy. The paper highlights the most effective approaches for leveraging deep learning in text classification and offers insights into future research directions in this domain.