<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0405">
  <Title>Using Summaries in Document Retrieval</Title>
  <Section position="11" start_page="0" end_page="0" type="ackno">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> Customers of online services approach their information seeking tasks from many perspectives, and yet most IR evaluations are conducted from a single user perspective. Using a single user perspective is easier, but it risks hiding the potential benefits of some new feature from key customer segments who might value it.</Paragraph>
    <Paragraph position="1"> News document summaries such as those in the form of Searchable LEAD provide a way to help some customer segments retrieve smaller answer sets that are focused on highly relevant documents. But the benefits of this are only apparent when the results are evaluated from that perspective.</Paragraph>
    <Paragraph position="2"> It is not simply a trade-off between recall and precision, but one between recall and precision with respect to the definition of relevance that that different customer segments have. An answer set of ten documents that mention but do not comment on some topic may result in 100% precision for the all reference evaluation scope but 0% precision for the highly relevant reference evaluation scope. Customer segments can and do have such widely divergent views of relevance.</Paragraph>
    <Paragraph position="3"> This evaluation showed that a customer seeking to retrieve a few highly relevant documents about some topic would benefit from using Searchable LEAD, retrieving smaller answer sets and a higher proportion of highly relevant documents in that answer set. A customer wanting to retrieve all documents that refer to the topic should avoid Searchable LEAD and instead continue to use full-text search.</Paragraph>
  </Section>
class="xml-element"></Paper>