<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0404">
  <Title>Learning Subjective Nouns using Extraction Pattern Bootstrapping</Title>
  <Section position="6" start_page="0" end_page="0" type="relat">
    <SectionTitle>
5 Related Work
</SectionTitle>
    <Paragraph position="0"> Several types of research have involved document-level subjectivity classification. Some work identifies inflammatory texts (e.g., (Spertus, 1997)) or classifies reviews as positive or negative ((Turney, 2002; Pang et al., 2002)).</Paragraph>
    <Paragraph position="1"> Tong's system (Tong, 2001) generates sentiment timelines, tracking online discussions and creating graphs of positive and negative opinion messages over time. Research in genre classification may include recognition of subjective genres such as editorials (e.g., (Karlgren and Cutting, 1994; Kessler et al., 1997; Wiebe et al., 2001)).</Paragraph>
    <Paragraph position="2"> In contrast, our work classifies individual sentences, as does the research in (Wiebe et al., 1999). Sentence-level subjectivity classification is useful because most documents contain a mix of subjective and objective sentences. For example, newspaper articles are typically thought to be relatively objective, but (Wiebe et al., 2001) reported that, in their corpus, 44% of sentences (in articles that are not editorials or reviews) were subjective.</Paragraph>
    <Paragraph position="3"> Some previous work has focused explicitly on learning subjective words and phrases. (Hatzivassiloglou and McKeown, 1997) describes a method for identifying the semantic orientation of words, for example that beautiful expresses positive sentiments. Researchers have focused on learning adjectives or adjectival phrases (Turney, 2002; Hatzivassiloglou and McKeown, 1997; Wiebe, 2000) and verbs (Wiebe et al., 2001), but no previous work has focused on learning nouns. A unique aspect of our work is the use of bootstrapping methods that exploit extraction patterns. (Turney, 2002) used patterns representing part-of-speech sequences, (Hatzivassiloglou and McKeown, 1997) recognized adjectival phrases, and (Wiebe et al., 2001) learned N-grams. The extraction patterns used in our research are linguistically richer patterns, requiring shallow parsing and syntactic role assignment. null In recent years several techniques have been developed for semantic lexicon creation (e.g., (Hearst, 1992; Riloff and Shepherd, 1997; Roark and Charniak, 1998; Caraballo, 1999)). Semantic word learning is different from subjective word learning, but we have shown that Meta-Bootstrapping and Basilisk could be successfully applied to subjectivity learning. Perhaps some of these other methods could also be used to learn subjective words.</Paragraph>
  </Section>
class="xml-element"></Paper>