<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0204">
  <Title>Improving Semi-Supervised Acquisition of Relation Extraction Patterns</Title>
  <Section position="9" start_page="33" end_page="33" type="evalu">
    <SectionTitle>
6 Results
</SectionTitle>
    <Paragraph position="0"> Results from the relation extraction evaluation can be seen in Table 2 and Figure 3. The seven seed patterns achieve a precision of 0.833 and recall of 0.022. The two approaches based on cosine similarity performs poorly, irrespective of the pattern model being used. The maximum increase in F-measure of 0.15 (when using the cosine measure with the linked chain model) results in a maximum F-measure for the cosine similarity model of 0.194 (with a precision of 0.491 and recall of 0.121) after 200 iterations.</Paragraph>
    <Paragraph position="1"> The best result is recorded when the linked chain model is used with the similarity measure introduced in Section 4.1, achieving a maximum F-measure of 0.329 (with a precision of 0.434 and recall of 0.265) after 190 iterations. This is not a high F-measure when compared against supervised IE systems, however it should be remembered that this represents an increase of 0.285 in F-measure over the original seven seed patterns and that this is achieved with a semi-supervised algorithm.</Paragraph>
  </Section>
class="xml-element"></Paper>