Automatic sentiment analysis of citizen comments: the case of the Albania earthquake

This study investigates how crowdsourced user comments collected through the LastQuake app can be used to improve situational awareness following an earthquake. Focusing on the 26 November 2019 Albania earthquake, the researchers analyzed a sample of 1,678 Albanian-language comments posted on the day of the event. Since such comments are unstructured and difficult to process manually during emergencies, the study explores the use of natural language processing techniques, specifically sentiment analysis, to automatically identify users' emotional responses and extract meaningful information from the data. 

The analysis revealed that negative sentiment was the most common polarity in the comments (52%), followed by positive and neutral sentiments. To assess whether sentiment classification could be automated effectively, the researchers evaluated two fine-tuned models, troberta and txlm, using manually classified comments as the reference standard. The results showed that troberta achieved an accuracy of 71%, significantly outperforming txlm, which reached 56% accuracy. These findings suggest that automatic sentiment analysis, particularly with the troberta model, has the potential to support rapid assessment of public reactions and needs during earthquake emergencies.