Improving sentiment analysis of disaster related social media content

Social media platforms have become the most accessible public communication and broadcast channels. Recently, the world has witnessed the prevailing usage of social media for communication during disasters. Being able to monitor and predict public opinions on social media during disasters allows us to evaluate crisis communication theories in order to design more efficient and effective communication mechanisms during the crisis. However, this potential is yet to be materialized due to difficulties in sentiment analysis of social media content. We propose to augment the effectiveness of such analysis by incorporating social relations in sentiment classification models. This thesis extends previous work substantially by looking at social relations of different nature, focusing on different communication goals at each stage of disaster management. This study provides a quantitative analysis of social media sentiments during disaster utilizing improved sentiment analysis and feature extraction techniques.