<?xml version="1.0" standalone="yes"?>
<Paper uid="I05-2027">
  <Title>Machine Learning Approach To Augmenting News Headline Generation</Title>
  <Section position="8" start_page="158" end_page="158" type="evalu">
    <SectionTitle>
ROUGE-L and ROUGE-W scores worked best
</SectionTitle>
    <Paragraph position="0"> As the results show the best performing topic labeling techniques are the TF and Hybrid systems. TF system is a baseline system that chooses high frequency content words as topic descriptors. Hybrid system is our decision tree classifier described in the previous section.</Paragraph>
    <Paragraph position="1"> Both of these systems outperform the Topiary's UTD method. The top three performing systems in this table combine topic labels with a compressed version of the lead sentence. Comparing these results to the Trim system (that returns the reduced lead sentence only), it is clear that the addition of topic descriptors greatly improves summary quality.</Paragraph>
    <Paragraph position="2"> The performance of the baseline TFTrim system and the HybridTrim system are very similar for all Rouge metrics; however, both systems outperform the Topiary headline generator.</Paragraph>
  </Section>
class="xml-element"></Paper>