Transformer-Based Multi-Class News Classification and Topic Detection

Abstract

The objective of this research is to enhance the implementation of transformer-based models for the classification of news and the identification of subjects on digital news platforms. This paper examines the application of sophisticated deep learning architectures, such as BERT, RoBERTa, and DistilBERT, to the classification of news and the identification of topics in large text corpus datasets. The semantic links and context of news articles may be difficult for machine learning algorithms to comprehend. Attention techniques are employed by transformer models to comprehend intricate linguistic patterns. The proposed system's data preprocessing, tokenization, feature extraction, model training, and evaluation are assessed using the F1-score, recall, precision, and accuracy. Standard algorithms are outperformed by transform-based approaches in the recognition and categorization of subjects, as demonstrated by experiments. This work facilitates the development of intelligent news analysis systems in contemporary media applications by enabling the categorization of content, the eradication of fake content, the provision of personalized recommendations, and the administration of real-time information.