Context enhanced word vectors for sentiment analysis

Abstract

Word vectors have become very important features for sentiment analysis. The aim of this paper is to encode sentimental context into pre-trained word vectors for sentiment analysis. The negation and intensity words in a context, as well as the sentimental words are combined to form context enhanced word vectors. Experiments on the datasets of SemEval show that the method of using intensity words has improved the result comparing with the baseline. The context enhanced words vectors from a distant supervision data can increase the similarity of the same polarity and decrease the similarity of the different polarity.