Sentiment classification of Chinese online reviews: a comparison of factors influencing performances

With the growing availability and popularity of online consumer reviews, people have been trying to seek sentiment-aware applications to gather and understand these opinion-rich texts. Thus, sentiment classification arises in response to analyse opinions of others automatically. In this paper, experiments of sentiment classification of Chinese online reviews across different domains are conducted by considering a couple of factors which potentially influence the sentiment classification performance. Experimental results indicate that the size of training sets and the number of features have certain influence on classification accuracy. In addition, there is no significant difference in classification accuracy when using Document Frequency, Chi-square Statistic and Information Gain, respectively, to reduce dimensionality. Low-order n-grams outperforms high-order n-grams in terms of accuracy if n-grams is taken as features. Furthermore, when words and combination of words are selected as features, the accuracy of adjectives is much close to that of NVAA (the combination of nouns, verbs, adjectives and adverbs), and is better than others as well.