<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2074">
  <Title>ARE: Instance Splitting Strategies for Dependency Relation-based Information Extraction</Title>
  <Section position="9" start_page="577" end_page="577" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> The current state-of-art IE methods tend to use co-occurrence relations for extraction of entities. Although context may provide a meaningful clue, the use of co-occurrence relations alone has serious limitations because of alignment and paraphrasing problems. In our work, we proposed to utilize dependency relations to tackle these problems. Based on the extracted anchor cues and relations between them, we split instances into 'simple', 'average' and 'hard' categories. For each category, we applied specific strategy. This approach allowed us to outperform the existing state-of-art approaches by 3% on Terrorism domain and 6% on Management Succession domain. In our future work we plan to investigate the role of semantic relations and integrate ontology in the rule generation process. Another direction is to explore the use of bootstrapping and transduction approaches that may require less training instances.</Paragraph>
  </Section>
class="xml-element"></Paper>