Enhancing Negation-Aware Sentiment Classification on Product Reviews via Multi-Unigram Feature Generation

摘要

Sentiment classification on product reviews has become a popular topic in the research community. In this paper, we propose an approach to generating multi-unigram features to enhance a negation-aware Naive Bayes classifier for sentiment classification on sentences of product reviews. We coin the term "multi-unigram feature" to represent a new kind of features that are generated in our proposed algorithm with capturing high-frequently co-appeared unigram features in the training data. We further make the classifier aware of negation expressions in the training and classification process to eliminate the confusions of the classifier that is caused by negation expressions within sentences. Extensive experiments on a human-labeled data set not only qualitatively demonstrate good quality of the generated multi-unigram features but also quantitatively show that our proposed approach beats three baseline methods. Experiments on impact analysis of parameters illustrate that our proposed approach stably outperforms the baseline methods.