Classification of Eyewitness Tweets in Emergency Situations.

Social platforms such as Twitter offer important information about disasters in emergency situations. Unique information found on these platforms is provided by people directly involved. Presenting these data to rescue teams can make a significant difference in how the situation is managed and how resources are distributed. Identification of relevant tweets can be done with Machine Learning and Natural Language Processing techniques. Various supervised and unsupervised learning algorithms have been previously used for this problem, including diverse heuristics. The purpose of this project is to explore and compare several approaches, test variations of parameters, and filter input data in order to improve performance. Challenges posed by the class imbalance present in emergency situations and the language diversity on social media platforms are also discussed.