<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0404">
  <Title>Learning Subjective Nouns using Extraction Pattern Bootstrapping</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This research produced interesting insights as well as performance results. First, we demonstrated that weakly supervised bootstrapping techniques can learn subjective terms from unannotated texts. Subjective features learned from unannotated documents can augment or enhance features learned from annotated training data using more traditional supervised learning techniques. Second, Basilisk and Meta-Bootstrapping proved to be useful for a different task than they were originally intended.</Paragraph>
    <Paragraph position="1"> By seeding the algorithms with subjective words, the extraction patterns identified expressions that are associated with subjective nouns. This suggests that the bootstrapping algorithms should be able to learn not only general semantic categories, but any category for which words appear in similar linguistic phrases. Third, our best subjectivity classifier used a wide variety of features. Subjectivity is a complex linguistic phenomenon and our evidence suggests that reliable subjectivity classification requires a broad array of features.</Paragraph>
  </Section>
class="xml-element"></Paper>