Comparative opinion sentences identification and elements extraction

This paper presents a subsystem for analyzing the comparative opinion sentences. This subsystem consists of three modules. Firstly, rule-based and CSR-based method is developed to identify comparative sentences, respectively. Secondly, in opinion element extraction module, CRFs model is applied to label elements. Finally, in the polarity determination module, some heuristic rules are complied to classify different types of comparative sentences. The evaluations on COAE2012 dataset show that the proposed analysis subsystem achieves a good result.