<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0204">
  <Title>Improving Semi-Supervised Acquisition of Relation Extraction Patterns</Title>
  <Section position="10" start_page="33" end_page="33" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> A number of conclusions can be drawn from the work described in this paper. Firstly, semi-supervised approaches to IE pattern acquisition benefit from the use of more expressive extraction pattern models since it has been shown that the performance of the linked chain model on the relation extraction task is superior to the simpler SVO model. We have previously presented a theoretical analysis (Stevenson and Greenwood, 2006a) which suggested that the linked chain model was a more suitable format for IE patterns than the SVO model but these experiments are, to our knowledge, the first to show that applying this model improves learning performance. Secondly, these experiments demonstrate that similarity measures inspired by kernel functions developed for use in supervised learning algorithms can be applied to semi-supervised approaches. This suggests that future work in this area should consider applying other similarity functions, including kernel methods, developed for supervised learning algorithms to the task of semi-supervised IE pattern acquisition. Finally, we demonstrated that this similarity measure outperforms a previously proposed approach which was based on cosine similarity and a vector space representation of patterns (Stevenson and Greenwood, 2005).</Paragraph>
  </Section>
class="xml-element"></Paper>