An Iterative Reinforcement Approach for Fine-Grained Opinion Mining

Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the ACL, pages 486–493, Boulder, Colorado, June 2009. c⃝2009 Association for Computational Linguistics An Iterative Reinforcement Approach for Fine-Grained Opinion Mining Weifu Du Haerbin Institute of Technology Haerbin, China duweifu@software.ict.ac.cn Songbo Tan Institute of Computing Technology Beijing, China tansongbo@software.ict.ac.cn Abstract With the in-depth study of sentiment analysis research, finer-grained opinion mining, which aims to detect opinions on different review fea- tures as opposed to the whole review level, has been receiving more and more attention in the sentiment analysis research community re- cently. Most of existing approaches rely mainly on the template extraction to identify the ex- plicit relatedness between product feature and opinion terms, which is insufficient to detect the implicit review features and mine the hid- den sentiment association in reviews, which satisfies (1) the review features are not appear explicit in the review sentences; (2) it can be deduced by the opinion words in its context. From an information theoretic point of view, this paper proposed an iterative reinforcement framework based on the improved information bottleneck algorithm to address such problem. More specifically, the approach clusters prod- uct features and opinion words simultaneously and iteratively by fusing both their semantic in- formation and co-occurrence information. The experimental results demonstrate that our ap- proach outperforms the template extraction based approaches. 1 Introduction In the Web2.0 era, the Internet turns from a static information media into a platform for dynamic information exchanging, on which people can ex- press their views and show their selfhood. More and more people are willing to record their feel- ings (blog), give voice to public affairs (news re- view), express their likes and dislikes on products (product review), and so on. In the face of the vol- ume of sentimental information available on the Internet continues to increase, there is growing interest in helping people better find, filter, and manage these resources. Automatic opinion mining (Turney et al., 2003; Ku et al., 2006; Devitt et al., 2007) can play an important role in a wide variety of more flexible and dynamic information management tasks. For example, with the help of sentiment analysis sys- tem, in the field of public administration, the ad- ministrators can receive the feedbacks on one pol- icy in a timelier manner; in the field of business, manufacturers can perform more targeted updates on products to improve the consumer experience. The research of opinion mining began in 1997, the early research results mainly focused on the polarity of opinion words (Hatzivassiloglou et al., 1997) and treated the text-level opinion mining as a classification of either positive or negative on the number of positive or negative opinion words in one text (Turney et al., 2003; Pang et al., 2002; Zagibalov et al., 2008;). With the in-depth study of opinion mining, researchers committed their ef- forts for more accurate results: the research of sen- timent summarization (Philip et al., 2004; Hu et al., KDD 2004), domain transfer problem of the …

Content truncated to 500 words