A Study of Dependency Features for Chinese Sentiment Classification

摘要: Syntactic dependency features, which encode long-range relations and word order information, have been employed in sentiment classification. However, much of the research has done English, researches conducted on exploring how features based syntactic can be utilized Chinese classification are very rare. In this study, we present an empirical study for First, consider two types feature sets (word unigrams word-dependency relations), three commonly-used weighting schemes (term presence, term frequency, TF-IDF), well-known learning methods (Naive Bayes SVM) to evaluate performance different classifiers. Then, use ensemble technique combine algorithms. Specifically, methods, namely average combination method meta-learning method, evaluated strategies. Through a wide range comparative experiments widely-used datasets classification, finally, some in-depth discussion is presented conclusions drawn about effectiveness