OBJECTIVE: The aim of this study is to develop disaster specific classification model specific to India, that extracts population emotional states during disasters from geo-referenced social media data. The developed model is also examined considering the case of Chennai floods 2015. METHODS: Using tweets from various disaster situations in India, a domain-specific training corpus was created. After training and testing different machine learning models, the top-performing model was used to classify emotions from tweets collected between November and December 2015 from Chennai. Using QGIS and Z-proportional testing, the spatial and temporal variations of negative emotions were further investigated. RESULTS: Emotions were extracted from geo-tagged tweets that were posted during the 2015 Chennai floods using the SVM classifier (accuracy: 0.84). The findings revealed that, in comparison to before the flood, there was a significant increase in sadness (23%) and fear (6%) during the disaster. Sadness and disgust decreased in post flood period. Distinct spatiotemporal patterns of emotional expression were identified, which showed clusters of anger in Perungudi, fear in Adyar, and grief and disgust in Teynampet and Sholinganallur zones. CONCLUSION: This study remains a proof-of-concept confirming the importance of social media analytics during disasters for directing focused response and recovery activities.
Karmegam et al. (Thu,) studied this question.