Sentiment Classification of Short Texts

Abstract

Over the few years, Sentiment analysis has been the heart of social media research due to the huge volume of opinionated data available on the web and its pervasive real life and commercial applications. Sentiment classification of shorter texts such as movie reviews is challenging due to lack of contextual information which often leads to interesting and unexpected results. Historically, this problem has been addressed using machine learning algorithms that usually learn from rule-based approaches or manually defined sparse features. In the recent years, Deep Neural Networks have gained a lot of attention in sentiment analysis due to their ability to effectively capture subtle semantic information from the input. These methods are capable of building dense continuous feature vectors, which is difficult to model in conventional models such as bag-of-words. In this paper, we conduct experiments and compare several machine learning algorithms Support Vector Machine, Naive Bayes, Random Forest, and a Deep Learning Algorithm. We selected Convolution Neural Network (CNN) trained on top of various pre-trained word vectors for movie review classification. We validate above models on IMDB movie review dataset, experimental results demonstrate that the task of sentiment analysis can benefit more from the CNN rather than the machine learning techniques.