A Subword-Based Deep Learning Approach for Sentiment Analysis of Political Tweets

Abstract

The successful use of online material in political campaigns over the past two decades has motivated the inclusion of social media platforms—such as Twitter—as an integral part of the political apparatus. Political analysts are increasingly turning to Twitter as an indicator of public opinion. We are interested in learning how positive and negative opinions propagate through Twitter and how important events influence public opinion. In this paper, we present a neural network-based approach to analyse the sentiment expressed on political tweets. First, our approach represents the text by dense vectors comprising subword information to better detect word similarities by exploiting both morphology and semantics. Then, a Convolutional Neural Network is trained to learn how to classify tweets depending on sentiment, based on an available labelled dataset. Finally, the model is applied to perform the sentiment analysis of a collection of tweets retrieved during the days prior to the latest UK General Election. Results are promising and show that the neural network approach represents an improvement over lexicon-based approaches for positive/negative sentence classification.