Comparison is widely used by consumers during the process of product evaluation in order to emphasize their preference, which can contribute to a proxy for product competitiveness analysis. This paper proposes a novel method for mining comparative sentences based on the achievements of linguistic study. The definition of comparative sentence subcategory is put forward and a mixed rule pool containing both artificial rules and CSR is set up. Besides, an entity dictionary is used to re-check the identification result which can ensure precise identification and classification of comparative sentences. Real online comments are collected from Dianping.com as experimental data. The result shows that the proposed method outperforms baseline methods in terms of identification precision. Based on the result, features and opinions of comparative sentences are mined. We then conducted sentiment analysis to calculate the sentimental score of comparison relations. Finally, a feature competitive network of restaurants is constructed.