<?xml version="1.0" standalone="yes"?>
<Paper uid="I05-2011">
  <Title>Automatic Detection of Opinion Bearing Words and Sentences</Title>
  <Section position="7" start_page="65" end_page="65" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper, we described an efficient automatic algorithm to produce opinion-bearing words by combining two methods. The first method used only a small set of human-annotated data. We showed that one can find productive synonyms and antonyms of an opinion-bearing word through automatic expansion in WordNet and use them as feature sets of a classifier. To determine a word's closeness to opinion-bearing or non-opinion-bearing synoym set, we also used all synonyms of a given word as well as the word itself. An additional method, harvesting words from WSJ, can compensate the first method.</Paragraph>
    <Paragraph position="1"> Using the resulting list, we experimented with different cutoff thresholds in the opinion/nonopinion sentence classification on 3 different test data sets. Especially on the TREC 2003 Novelty Track, the system performed well. We plan in future work to pursue the automated analysis of exhortatory text in order to produce detailed argument graphs reflecting their authors' argumentation.</Paragraph>
  </Section>
class="xml-element"></Paper>