Extracting Resource Terms for Sentiment Analysis

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1171–1179, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Extracting Resource Terms for Sentiment Analysis Lei Zhang Department of Computer Science University of Illinois at Chicago 851 S. Morgan St, Chicago IL 60607 lzhang3@cs.uic.edu Bing Liu Department of Computer Science University of Illinois at Chicago 851 S. Morgan St, Chicago IL 60607 liub@cs.uic.edu Abstract Existing research on sentiment analysis mainly uses sentiment words and phrases to determine sentiments expressed in documents and sen- tences. Techniques have also been developed to find such words and phrases using dictiona- ries and domain corpora. However, there are still other types of words and phrases that do not bear sentiments on their own, but when they appear in some particular contexts, they imply positive or negative opinions. One class of such words or phrases is those that express resources such as water, electricity, gas, etc. For example, “this washer uses a lot of elec- tricity” is negative but “this washer uses little water” is positive. Extracting such resource words and phrases are important for sentiment analysis. This paper formulates the problem based on a bipartite graph and proposes a nov- el iterative algorithm to solve the problem. Experimental results using diverse real-life sentiment corpora show good results. 1 Introduction Sentiment analysis or opinion mining has been an active research area in recent years (e.g., Pang and Lee 2008; Turney, 2002; Wiebe et al. 2004; Hu and Liu, 2004; Kim and Eduard, 2004; Wil- son et al. 2005; Popescu and Etzioni, 2005; Ri- loff et al. 2006; Esuli and Fabrizio, 2006; Mei et al, 2007; Stoyanov and Cardie; 2008). Research- ers have studied the problem at the document level, sentence level and aspect level to deter- mine the sentiment polarity expressed in a doc- ument, in a sentence and on an aspect of an enti- ty (see the surveys (Pang and Lee, 2008) and (Liu, 2010)). One type of key information used in almost all existing sentiment analysis tech- niques is a list of sentiment words (or opinion words). Positive sentiment words are words ex- pressing desired states or qualities, e.g., good, amazing, and excellent, and negative sentiment words are words expressing undesirable states or qualities, e.g., bad, crappy, and ugly. A key characteristic of these words is that they themselves bear sentiments. They are frequently used in sentiment analysis tasks. However, it is also important to recognize that sentiment analy- sis based only on these words (or phrases) is far from sufficient. There are still many other types of expressions that do not bear sentiments on their own, but when they appear in some particu- lar contexts, they imply sentiments. In (Liu, 2010), several such expressions and their corres- ponding opinion/sentiment rules are introduced. We believe that all these expressions have to be extracted and associated problems solved before sentiment analysis can achieve the next level of accuracy. One such type of expressions involves resources, which occur frequently in many …

Content truncated to 500 words