Multilingual Sentiment Analysis on Social Media Disaster Data

The use of social media in disaster situations is inevitable, but it is also the case that information presented through this medium can include both public opinion and general information. In a multicultural nation like Malaysia, people like to use codeswitch sentences, through which they mix several languages to express their opinions. Sentiment analysis can be used to classify the subjectivity of social media data, by considering the multilingual aspect of Malaysian users who may experience disaster. In this paper, the authors propose a multilingual sentiment classifier used to understand how Malaysians react during a disaster. The proposed model collects disaster data from social media, which is then classified through a deep learning algorithm, so as to analyze the sentiments of people affected by disasters. The experiment results show that a multilingual sentiment classifier can achieve 0.862 accuracy and 0.864 F1-score which is considered suitable for analyzing social media data. The classification result shows that most Malaysians use social media to disseminate information during disaster periods.