A Cross-Analyzing Approach to Sentiment and Bias Detection in Social Media: Insights from Geopolitical Conflicts

Sentiment analysis has become a critical tool for analyzing public opinion, especially in the context of geopolitical conflicts like the Russia-Ukraine war and the Gaza-Israel conflict. This paper leverages advanced Natural Language Processing (NLP) models, namely Long Short-Term Memory (LSTM) networks and Bidirectional Encoder Representations from Transformers (BERT), to classify sentiments and detect biases in social media content. By utilizing feature engineering techniques such as BERT tokenization and GloVe embeddings, the models are trained to capture the nuances of public discourse. The experimental results show that BiLSTM achieves a sentiment classification accuracy of 99.28%, slightly outperforming BERT (99.27%). For bias detection, BERT demonstrates superior performance with an accuracy of 80%, effectively identifying pro-Israel, pro-Palestine, and neutral biases. The study highlights the relationship between sentiment and bias, revealing that neutral sentiments are strongly correlated with neutral biases, while positive sentiments align more with pro-Palestine bias. Despite the model’s effectiveness, challenges remain in handling class imbalances and differentiating between subtle biases. These findings provide valuable insights for media analysis, policymaking, and the future development of more robust sentiment and bias detection systems.