<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1026">
  <Title>The Effectiveness of Dictionary and Web-Based Answer Reranking</Title>
  <Section position="8" start_page="10" end_page="10" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> We described dictionary-based answer reranking using WordNet, web-based answer reranking using three different online search engines, and their evaluations at various parameter settings on a set of 102 TREC-10 definition questions. We showed that using either approach alone improved MRR score by 19% and PCT5 score by 5% over the baseline. However, the best performance was achieved when both methods were used together. In that setting a 25% increase in MRR score and 14% improvement in PCT5 score were obtained.</Paragraph>
    <Paragraph position="1"> The difference on the best MRR and PCT5 scores (0.56 vs. 0.73) suggests neither dictionary-based nor web-based will solve the reranking problem completely.</Paragraph>
    <Paragraph position="2"> To improve the performance further, we need better ways to compile web glosses and combine them with WordNet glosses. We also need a better combination function a statistical model for combining patterns, dictionary, and web scores. We have started investigating the possibility of applying answer reranking to other question types and exploring specialized web resources.</Paragraph>
  </Section>
class="xml-element"></Paper>