Multitask Sentiment Analysis and Topic Classification Using BERT

Abstract

In this study, a multitask model is proposed to perform simultaneous news category and sentiment classification of a diverse dataset comprising 3263 news records spanning across eight categories, including environment, health, education, tech, sports, business, lifestyle, and science. Leveraging the power of Bidirectional Encoder Representations from Transformers (BERT), the algorithm demonstrates remarkable results in both tasks. For topic classification, it achieves an accuracy of 98% along with balanced precision and recall, substantiating its proficiency in categorizing news articles. For sentiment analysis, the model maintains strong accuracy at 94%, distinguishing positive from negative sentiment effectively. This multitask approach showcases the model's versatility and its potential to comprehensively understand and classify news articles based on content and sentiment. This multitask model not only enhances classification accuracy but also improves the efficiency of handling extensive news datasets. Consequently, it empowers news agencies, content recommendation systems, and information retrieval services to offer more personalized and pertinent content to their users.