<?xml version="1.0" standalone="yes"?>
<Paper uid="X98-1025">
  <Title>SUMMARIZATION: (1) USING MMR FOR DIVERSITY- BASED RERANKING AND (2) EVALUATING SUMMARIES</Title>
  <Section position="12" start_page="193" end_page="194" type="concl">
    <SectionTitle>
10. CONCLUSION
</SectionTitle>
    <Paragraph position="0"> We have shown that MMR ranking provides a useful and beneficial manner of providing information to the user by allowing the user to minimize redundancy. This is especially true in the case of query-relevant multi-document summarization in this one data collection. We are currently performing studies on how this extends to additional document collections. In the future we will also be investigating how to handle co-reference in our system as well as analyzing the most suitable ~, par/maeters and clustering the output results.</Paragraph>
    <Paragraph position="1"> Text Summarization is still in the infant stage in terms of evaluation. Many monolingual document information retrieval results can be applied to text summarization, but as of yet, there has been little evaluation of these techniques. This pilot experiment showed many areas that need to be examined in further detail, including whether the summary selects the most relevant sentences in the document and whether these results generalize to more data sets and other document genres. We also plan to explore further the effects of query expansion using WordNet, as well as the use the first sentence (for news stories) in the query and/or summary. We also plan to run experiments fixing the number of sentences for each document as the number of relevant sentences chosen by the assessors as well as a small number, such as three. We are currently in the process of building a more extensive sentence relevance database for further evaluation. In this database, we are collecting data on the user selected most relevant sentence(s) for each document. We also plan to explore how to join the relevant sections to provide a &amp;quot;good&amp;quot;, understandable, readable, relevant, non-redundant summary.</Paragraph>
  </Section>
class="xml-element"></Paper>