Resolving Object and Attribute Coreference in Opinion Mining

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 268–276, Beijing, August 2010 Resolving Object and Attribute Coreference in Opinion Mining Xiaowen Ding Department of Computer Science University of Illinois at Chicago dingxwsimon@gmail.com Bing Liu Department of Computer Science University of Illinois at Chicago liub@cs.uic.edu Abstract Coreference resolution is a classic NLP problem and has been studied extensively by many researchers. Most existing studies, however, are generic in the sense that they are not focused on any specific text. In the past few years, opinion mining became a popular topic of research because of a wide range of applications. However, limited work has been done on coreference resolu- tion in opinionated text. In this paper, we deal with object and attribute coreference resolution. Such coreference resolutions are important because without solving it a great deal of opinion information will be lost, and opinions may be assigned to wrong entities. We show that some important features re- lated to opinions can be exploited to perform the task more accurately. Experimental re- sults using blog posts demonstrate the effec- tiveness of the technique. 1 Introduction Opinion mining has been actively researched in recent years. Researchers have studied the prob- lem at the document level (e.g., Pang et al., 2002; Tuney, 2002; Gamon et al., 2005) sen- tence and clause level (Wilson et al., 2004; Kim and Hovy, 2004), word level (e.g., Andreevs- kaia and Bergler, 2006; Hatzivassiloglou and McKeown, 1997; Esuli and Sebastiani, 2006; Kanayama and Nasukawa, 2006; Qiu et al., 2009), and attribute level (Hu and Liu 2004; Popescu and Etzioni, 2005; Ku et al., 2006; Mei et al., 2007; Titov and McDonald 2008). Here attributes mean different aspects of an object that has been commented on. Let us use the fol- lowing example blog to illustrate the problem: “I bought a Canon S500 camera yesterday. It looked beautiful. I took a few photos last night. They were amazing”. “It” in the second sen- tence refers to “Canon S500 camera”, which is called an object. “They” in the fourth sentence refers to “photos”, which is called an attribute of the object “Canon S500 camera”. The use- fulness of coreference resolution in this case is clear. Without resolving them, we lose opinions. That is, although we know that the second and fourth sentences express opinions, we do not know on what. Without knowing the opinion target, the opinion is of limited use. In (Nicolov et al., 2008), it was shown based on manually annotated data that opinion mining results can be improved by 10% if coreference resolution is used (the paper did not provide an algorithm). In this paper, we propose the problem of ob- ject and attribute coreference resolution – the task of determining which mentions of objects and attributes refer to the same entities. Note that here entities refer to both objects and attributes, not the traditional named entities. To our knowledge, limited work has been done on this problem in the opinion mining context apart from …

Content truncated to 500 words