A Comparative Analysis of Deep Learning Methods Across Time in Filtering Unadulterated Positive News Articles

The media today is sensationalized. It is often skewed towards the negative side of the world to prompt a reaction out of its viewers and keep them engaged. This method of content curation increases stress, anxiety, and promotes a negative outlook on the state of the world. Traditional news platforms thus prioritize content that captures attention, overlooking the more uplifting and positive stories happening around us. To solve this problem, we explore various sentiment analysis techniques to filter and deliver uplifting news. The study aims to analyse the difference in efficacy between a more modern transformer-based approach, specifically the Robustly Optimized BERT-Pretraining Approach (RoBERTa), and a more conventional Feed Forward Neural Network (FFNN) trained using Term Frequency – Inverse Document Frequency (TF-IDF). We aim to demonstrate the advancements in deep learning over time by evaluating models using criteria such as accuracy and F1-score. The findings highlight how modern NLP techniques can enhance sentiment-based news filtering, spotlighting the evolution of deep learning techniques in permitting personalized and positive content curation.