Clustering Aspect-related Phrases by Leveraging Sentiment Distribution Consistency

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1614–1623, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Clustering Aspect-related Phrases by Leveraging Sentiment Distribution Consistency Li Zhao, Minlie Huang, Haiqiang Chen*, Junjun Cheng*, Xiaoyan Zhu State Key Laboratory of Intelligent Technology and Systems National Laboratory for Information Science and Technology Dept. of Computer Science and Technology, Tsinghua University, Beijing, PR China *China Information Technology Security Evaluation Center zhaoli19881113@126.com aihuang@tsinghua.edu.cn Abstract Clustering aspect-related phrases in terms of product’s property is a precursor pro- cess to aspect-level sentiment analysis which is a central task in sentiment analy- sis. Most of existing methods for address- ing this problem are context-based models which assume that domain synonymous phrases share similar co-occurrence con- texts. In this paper, we explore a novel idea, sentiment distribution consistency, which states that different phrases (e.g. “price”, “money”, “worth”, and “cost”) of the same aspect tend to have consistent sentiment distribution. Through formal- izing sentiment distribution consistency as soft constraint, we propose a novel unsu- pervised model in the framework of Poste- rior Regularization (PR) to cluster aspect- related phrases. Experiments demonstrate that our approach outperforms baselines remarkably. 1 Introduction Aspect-level sentiment analysis has become a cen- tral task in sentiment analysis because it can ag- gregate various opinions according to a product’s properties, and provide much detailed, complete, and in-depth summaries of a large number of re- views. Aspect finding and clustering, a precursor process of aspect-level sentiment analysis, has at- tracted more and more attentions (Mukherjee and Liu, 2012; Chen et al., 2013; Zhai et al., 2011a; Zhai et al., 2010). Aspect finding and clustering has never been a trivial task. People often use different words or phrases to refer to the same product property (also called product aspect or feature in the literature). Some terms are lexically dissimilar while seman- tically close, which makes the task more challeng- ing. For example, “price”, “money” , “worth” and “cost” all refer to the aspect “price” in reviews. In order to present aspect-specific summaries of opinions, we first of all, have to cluster different aspect-related phrases. It is expensive and time- consuming to manually group hundreds of aspect- related phrases. In this paper, we assume that the aspect phrases have been extracted in advance and we keep focused on clustering domain synony- mous aspect-related phrases. Existing studies addressing this problem are mainly based on the assumption that different phrases of the same aspect should have similar co- occurrence contexts. In addition to the traditional assumption, we develop a new angle to address the problem, which is based on sentiment distribution consistency assumption that different phrases of the same aspect should have consistent sentiment distribution, which will be detailed soon later. Figure 1: A semi-structured Review. This new angle is inspired by this simple obser- vation (as illustrated in Fig. 1): two phrases within the same cluster are not likely to be simultaneously placed in Pros and Cons of the same review. A …

Content truncated to 500 words