Collocation Polarity Disambiguation Using Web-based Pseudo Contexts

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 160–170, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Collocation Polarity Disambiguation Using Web-based Pseudo Contexts Yanyan Zhao, Bing Qin and Ting Liu∗ Harbin Institute of Technology, Harbin, China {yyzhao, bqin, tliu}@ir.hit.edu.cn Abstract This paper focuses on the task of colloca- tion polarity disambiguation. The collocation refers to a binary tuple of a polarity word and a target (such as ⟨long, battery life⟩ or ⟨long, startup⟩), in which the sentiment orientation of the polarity word (“long”) changes along with different targets (“battery life” or “startup”). To disambiguate a collocation’s polarity, pre- vious work always turned to investigate the polarities of its surrounding contexts, and then assigned the majority polarity to the collo- cation. However, these contexts are limited, thus the resulting polarity is insufficient to be reliable. We therefore propose an unsuper- vised three-component framework to expand some pseudo contexts from web, to help dis- ambiguate a collocation’s polarity.Without us- ing any additional labeled data, experiments show that our method is effective. 1 Introduction In recent years, more attention has been paid to sentiment analysis as it has been widely used in various natural language processing applications, such as question answering (Wiebe et al., 2003; Yu and Hatzivassiloglou, 2003), information extrac- tion (Riloff et al., 2005) and opinion-oriented sum- marization (Hu and Liu, 2004; Liu et al., 2005). Meanwhile, it also brings us lots of interesting and challenging research topics, such as subjectivity analysis (Riloff and Wiebe, 2003), sentiment clas- sification (Pang et al., 2002; Kim and Hovy, 2005; ∗Correspondence author: tliu@ir.hit.edu.cn Wilson et al., 2009; He et al., 2011), opinion re- trieval (Zhang et al., 2007; Zhang and Ye, 2008; Li et al., 2010) and so on. One fundamental task for sentiment analysis is to determine the semantic orientations of words. For example, the word “beautiful” is positive, while “ugly” is negative. Many researchers have devel- oped several algorithms for this purpose and gener- ated large static lexicons of words marked with prior polarities (Hatzivassiloglou and McKeown, 1997; Turney et al., 2003; Esuli, 2008; Mohammad et al., 2009; Velikovich et al., 2010). However, there exist some polarity-ambiguous words, which can dynam- ically reflect different polarities along with different contexts. A typical polarity-ambiguous word “长” (“long” in English) is shown with two example sen- tences as follows. 1. 该相机的[电池寿命]t很[长]p。(Positive) Translated as: The [battery life]t of this camera is [long]p. (Positive) 2. 该相机的[启动时间]t很[长]p。(Negative) Translated as: This camera has [long]p [startup]t. (Negative) The phrases marked with p superscript are the polarity-ambiguous words, and the phrases marked with t superscript are targets modified by the polar- ity words. In the above two sentences, the sentiment orientation of the polarity word “长” (“long” in En- glish) changes along with different targets. When modifying the target “电池寿命” (“battery life” in English), its polarity is positive; and when modify- ing “启动时间” (“startup” in English), its polarity is 160 negative. In this paper, we especially …

Content truncated to 500 words