<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3805">
  <Title>A Study of Two Graph Algorithms in Topic-driven Summarization</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Related work
</SectionTitle>
    <Paragraph position="0"> Erkan and Radev (2004), Mihalcea (2004), Mihalcea and Tarau (2004) introduced graph methods for summarization, word sense disambiguation and other NLP applications.</Paragraph>
    <Paragraph position="1"> The summarization graph-based systems implement a form of sentence ranking, based on the idea of prestige or centrality in social networks. In this case the network consists of sentences, and signi cantly similar sentences are interconnected. Various measures (such as node degree) help nd the most central sentences, or to score each sentence.</Paragraph>
    <Paragraph position="2"> In topic-driven summarization, one or more sentences or questions describe an information need which the summaries must address. Previous systems extracted key words or phrases from topics and used them to focus the summary (Fisher et al., 2005).</Paragraph>
    <Paragraph position="3"> Our experiments show that there is more to topics than key words or phrases. We will experiment with using grammatical dependency relations for the task of extractive summarization.</Paragraph>
    <Paragraph position="4"> In previous research, graph-matching using grammatical relations was used to detect textual entailment (Haghighi et al., 2005).</Paragraph>
  </Section>
class="xml-element"></Paper>