Exploiting Multilingual Neural Linguistic Representation for Sentiment Classification of Political Tweets in Code-mix Language

Social media is more and more utilized by people around the world to express their feelings and opinions in the kind of short text messages. Twitter has been a rapidly growing microblogging social networking website where people express their opinions in a precise and simple manner of expressions. It has also become a platform for governmental, non-governmental and individual opinions and policy announcements. Detecting sentiments from tweets has a wide range of applications including identifying the anxiety or depression of individuals and measuring the well-being or mood of a community. In addition, the sentiment classification becomes complex when the tweet is written in codemix language which is a mix of two different languages. The main objective of this paper is to classify tweets written in mix of Tamil and English language into positive, negative, or neutral. This is achieved by fine tuning a pretrained multilingual text representation model as well as deep learning transformers. The proposed approach is experimented with large scale of tweets collected for societal issues in India. We also provide a comparative study of different machine learning and deep learning models. The proposed architecture based on neural linguistic representation provides significant accuracy in classifying both Tamil and codemix tweets.