Sentiment Analysis of Social Media Posts During Crisis Situations

Abstract

Social media platforms are becoming one of important tools for connecting people, with millions of users and an unbelievable amount of content, it's becoming big data store for automated analysis. During recent crises, people have turned to these platforms seeking information, support, or coordination. The objective of this thesis is to evaluate how to use sentiment analysis techniques on social media content during crisis events, see if social media can be used to improve the situational awareness of non-governmental organizations working to provide humanitarian aid. To verify that, a study of sentiment analysis approaches and machine learning methods is performed, and an implementation of selected ML models for sentiment analysis is developed.

Social media content is unstructured and requires significant effort to collect, preprocess and analyze, for this study, data from Ethiopia during Tigray war was collected and used as a test case, these data was then annotated and prepared for sentiment analysis. Two machine learning algorithms were developed and tested, and while the resulted performance was limited, mostly because of limitations of available training data, the approach can be improved and used to create more performant models that can satisfy the desired purpose.