<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3005">
  <Title>A Data Driven Approach to Relevancy Recognition for Contextual Question Answering</Title>
  <Section position="8" start_page="39" end_page="39" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper, we present a data driven approach, decision tree learning, for the task of relevancy recognition in contextual question answering. Experiments show that this approach achieves 93% accuracy on the TREC data, about 12% improvement from the rule-based algorithm reported by De Boni and Mananhar (2005). Moreover, this data driven approach requires much less human effort on investigating a specific data set and less human expertise to summarize rules from the observation. All the features we used in the training can be automatically extracted. This makes it straightforward to train a model in a new domain, such as the HandQA.</Paragraph>
    <Paragraph position="1"> Furthermore, decision tree learning is a white-box model and the trained tree is human interpretable. It shows that the path measure has the best information gain among the other semantic similarity measures.</Paragraph>
    <Paragraph position="2"> We also report our preliminary experiment results on context information fusion for question answering.</Paragraph>
  </Section>
class="xml-element"></Paper>