Words of War: A hybrid BERT-CNN approach for topic-wise sentiment analysis on The Russia-Ukraine War

Abstract

The Russia-Ukraine War has dramatically impacted the world, affecting economies, lives, and politics. The war is a common topic on social media, especially on platforms like YouTube. In this study, we analyzed YouTube comments from videos posted by popular news channels like CNN, BBC, etc., to understand people’s opinions about the war. We used a tool called VADER for sentiment analysis and an unsupervised BERT model to identify ten key topics related to the war, including humanitarian issues, economic challenges, political debates, and societal concerns. We then created a model that combines BERT’s ability to understand context with CNN’s feature extraction strengths. Unlike existing approaches, our model incorporates an extra input layer that considers the topic as a significant feature. This hybrid model effectively classifies sentiments with 92.26% accuracy. Our research provides insights into public perceptions and discussions about the Russia-Ukraine War, highlighting essential themes in the conversation.