<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3801">
  <Title>A Graphical Framework for Contextual Search and Name Disambiguation in Email</Title>
  <Section position="7" start_page="7" end_page="7" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have presented a scheme for representing a corpus of email messages with a graph of typed entities, and an extension of the traditional notions of document similarity to documents embedded in a graph.</Paragraph>
    <Paragraph position="1"> Using a boosting-based learning scheme to rerank outputs based on graph-walk related, as well as other domain-specific, features provides an additional performance improvement. The final results are quite strong: for the explored name disambiguation task, the method yields MAP scores in the mid-to-upper 80's. The person name identification task illustrates a key advantage of our approach--that context can be easily incorporated in entity disambiguation.</Paragraph>
    <Paragraph position="2"> In future work, we plan to further explore the scalability of the approach, and also ways of integrating this approach with language-modeling approaches for document representation and retrieval.</Paragraph>
    <Paragraph position="3"> An open question with regard to contextual (multisource) graph walk in this framework is whether it is possible to further focus probability mass on nodes that are reached from multiple source nodes. This may prove beneficial for complex queries.</Paragraph>
  </Section>
class="xml-element"></Paper>