Sentiment Polarity Detection in Bengali Tweets Using LSTM Recurrent Neural Networks

Abstract

A vast amount of user generated content in the form of blog posts, tweets and comments etc. are now available online. This huge amount of social media texts can be analyzed for evaluation of public opinions and deriving market intelligence. Sentiment analysis is the research area dealing with analysis of sentiment such as feelings, emotions and opinions expressed in the social media texts by the internet users. Sentiment polarity detection is an important sentiment analysis task. Sentiment polarity detection in tweets can be viewed as a task for classifying an input tweet into one of three categories -positive, negative or neutral. Now-a-days, a large number of Indian language social media texts are also becoming available on the web. In this paper, we present a sentiment polarity detection system for Bengali tweets that uses a special type of recurrent neural networks called LSTM. Our experiments using a Bengali tweet dataset show that our proposed system outperforms some existing systems used for sentiment polarity detection in Bengali tweets.