lavanya sk

Summarizing opinions from online reviews is an important and challenging task as customers make decisions about their purchase based on other user's opinions available in these reviews. The broad availability of portable devices such as mobile devices and tablets, condensing information for display on a relatively small screen has become a necessity for fine-grained summary. Feature based opinion summarization is such a fine-grained summarization that generates structured review summaries highlighting individual features of a product and users opinion expressed on them. Features and Feature-specific opinion words must be clustered simultaneously to provide more informative descriptions of the product. Co-clustering has emerged as an important technique for simultaneous clustering which uses co-occurrence frequency matrix while grouping. Unsupervised co-clustering models suffer from semantic redundancies i.e., reviewers may use different phrases to express their view on the same feature-opinion phrase. These phrases need to be grouped under same feature-opinion phrase in order to perform an effective clustering. To address this issue, recently constrained co-clustering approach has been proposed which incorporates prior knowledge in the form of constraints, in a semi-supervised setting. This paper proposes Sentic LDCC to solve this problem in a different setting by incorporating common sense knowledge in the form of seed words to shift clustering from syntactic to semantic level. Experimental results on feature-opinion phrase extracted from car reviews show that this method outperforms state-of-the-art models.