<?xml version="1.0" standalone="yes"?>
<Paper uid="N03-4006">
  <Title>QCS: A Tool for Querying, Clustering, and Summarizing Documents</Title>
  <Section position="4" start_page="0" end_page="0" type="evalu">
    <SectionTitle>
3 Results
</SectionTitle>
    <Paragraph position="0"> QCS was tested using data from the 2002 Document Understanding Conference (http://duc.nist.gov/), a conference focusing on summarization and the evaluation of summarization systems. The data consisted of 567 news articles categorized into four types, with one type consisting of articles covering a single natural disaster event reported within a seven day window.</Paragraph>
    <Paragraph position="1"> Results of one test producing 100-word extract summaries can be seen in Figure 1, where the query consisted of the words, &amp;quot;hurricane&amp;quot; and &amp;quot;earthquake&amp;quot;. The top three scoring clusters contained a total 55 articles (32, 11, and 12, respectively), producing the summaries shown in the figure. The topics of these three summaries were a hurricane near Jamaica, catastrophe insurance, and an earthquake in California, respectively. Despite the limitations of automatic summarization, this example illus- null 2) a navigation section with links to clustered documents (Q: top documents retrieved for the query and their scores, C: documents from which summary sentences were drawn and the sentence indices, S: links to multiple or single document summaries), and 3) an output viewing section, which here contains the default output of multiple document summaries for the topic clusters.</Paragraph>
    <Paragraph position="2"> trates the utility of summarizing by cluster rather than producing a single summary of the retrieved documents.</Paragraph>
    <Paragraph position="3"> Further results are planned for the demonstration, including results of using QCS against the data from the</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2003 Document Understanding Conference.
Acknowledgements
</SectionTitle>
      <Paragraph position="0"> The authors would like to thank C. David Levermore and William D. Dorland of the University of Maryland for their helpful remarks concerning the QCS system.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>