Multilingual Sentiment Analysis on Twitter Data Towards Enhanced Policy Making

Abstract

The great expansion in the usage and popularity of social media platforms during the last decades has led to the production of an enormous real-time volume of social texts and posts, including tweets, that are being produced by users. These collections of social data can be potentially useful and provide useful insights to policymakers to adjust new user-centric policies and regulations. However, extracting and analyzing valuable information and knowledge out of these data is a challenging task as concerns the high multilingualism that describes these data. Thus, both the research and the business communities focus on the utilization of multilingual approaches and solutions to enhance the policy making procedures. To investigate a portion of these challenges this research work performs a comparative analysis of two multilingual sentiment analysis approaches. In this context, three multilingual deep learning classifiers and a zero-shot classification approach were utilized and compared. Their comparison has unveiled insightful outcomes and has a two-fold interpretation. Multilingual deep learning classifiers that have pre-trained and evaluated in monolingual data achieve high performances and transfer inference when applied afterwards in multilingual data. However, the zero-shot classification approach fails to achieve high accuracies in monolingual data as in contrary to when applied on multilingual data.