Sentiment Lexical-Augmented Convolutional Neural Networks for Sentiment Analysis

Sentiment analysis is the method of analyzing a piece of text for human sentiment or emotions. In this paper, we consider the sentence-level sentiment classification task. Recently an end-to-end convolutional neural network has been proposed to predict sentiment polarity directly. However, this approach extracts features only based on input word embedding sets, and fails to make full use of a word's information. It ignores the sentiment polarity of words which has been proved beneficial to sentiment classification. In this paper, we propose a Sentiment lexical-Augmented Convolutional Neural Network (SCNN) for sentiment analysis. We learn sentiment embedding for each word based on SentiWordNet, a widely used sentiment-lexical resource. Then both word embedding and sentiment embedding are fed into a convolutional neural network classifier. Experimental results on benchmark tasks show that our approach outperforms baseline methods.