<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1129">
  <Title>Exploring Distributional Similarity Based Models for Query Spelling Correction</Title>
  <Section position="8" start_page="1029" end_page="1029" type="concl">
    <SectionTitle>
5 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We have presented novel methods to learn better statistical models for the query spelling correction task by exploiting distributional similarity information. We explained the motivation of our methods with the statistical evidence distilled from query log data. To evaluate our proposed methods, two probabilistic models that can take advantage of such information are investigated.</Paragraph>
    <Paragraph position="1"> Experimental results show that both methods can achieve significant improvements over their baseline settings.</Paragraph>
    <Paragraph position="2"> A subject of future research is exploring more effective ways to utilize distributional similarity even beyond query logs. Currently for low-frequency terms in query logs there are no reliable distribution similarity evidence available for them. A promising method of dealing with this in next steps is to explore information in the resulting page of a search engine, since the snippets in the resulting page can provide far greater detailed information about terms in a query.</Paragraph>
  </Section>
class="xml-element"></Paper>