Search

Comparative sentences in online reviews express substantial information that are of concern to customers and attract increasing attention from manufacturers and service providers. At present studies of Chinese Comparative Sentence Identification are based on Pattern Matching and Supervised Machine Learning Algorithm, in which its performance requires further improvement. Therefore, the present analysis aims to further identify the candidate sets of comparative sentences by unsupervised sentiment analysis, and then improve mining performance. First, we constructed a Chinese comparative pattern set, which was used to extract candidate comparative sentences from the corpus of online products reviews. Moreover, we set the score for the candidate sentences using sentiment analysis technique. The experiment determined the threshold of positive and negative affective means, ranging from 0 to 0.03. Experimental results on Chinese customer reviews show that the final F-score value increased to 87.54%. In addition, a significant difference was set at the 0.01 level, which demonstrates the effectiveness of the technique. The proposed unsupervised method is suitable for the changeable and the large quantity of network review mining. This study does not only meet the need to generalize across different products and various data sizes but also improves the performance in terms of identifying comparative sentences. [ABSTRACT FROM AUTHOR]