<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0302">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Toward Opinion Summarization: Linking the Sources</Title>
  <Section position="4" start_page="9" end_page="9" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Sentiment analysis has been a subject of much recent research. Several efforts have attempted to automatically extract opinions, emotions, and sentiment from text. The problem of sentiment extraction at the document level (sentiment classification) has been tackled as a text categorization task in which the goal is to assign to a document eitherpositive(&amp;quot;thumbsup&amp;quot;)ornegative(&amp;quot;thumbs down&amp;quot;) polarity (e.g. Das and Chen (2001), Pang et al. (2002), Turney (2002), Dave et al. (2003), Pang and Lee (2004)). In contrast, the problem of fine-grained opinion extraction has concentrated on recognizing opinions at the sentence, clause, or individual opinion expression level. Recent work has shown that systems can be trained to recognize opinions, their polarity, and their strength at a reasonable degree of accuracy (e.g. Dave et al. (2003), Riloff and Wiebe (2003), Bethard et al. (2004), Pang and Lee (2004), Wilson et al.</Paragraph>
    <Paragraph position="1"> (2004), Yu and Hatzivassiloglou (2003), Wiebe and Riloff (2005)). Additionally, researchers have been able to effectively identify sources of opinions automatically (Bethard et al., 2004; Choi et al., 2005; Kim and Hovy, 2005). Finally, Liu et al.</Paragraph>
    <Paragraph position="2"> (2005) summarize automatically generated opinions about products and develop interface that allows the summaries to be vizualized.</Paragraph>
    <Paragraph position="3"> Our work also draws on previous work in the area of coreference resolution, which is a relatively well studied NLP problem. Coreference resolution is the problem of deciding what noun phrases in the text (i.e. mentions) refer to the same real-world entities (i.e. are coreferent). Generally, successful approaches have relied machine learning methods trained on a corpus of documents annotated with coreference information (such as the MUC and ACE corpora). Our approach to source coreference resolution is inspired by the state-of-the-art performance of the method of Ng and Cardie (2002).</Paragraph>
  </Section>
class="xml-element"></Paper>