<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1665">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Context-Dependent Term Relations for Information Retrieval</Title>
  <Section position="6" start_page="558" end_page="558" type="concl">
    <SectionTitle>
6. Conclusions
</SectionTitle>
    <Paragraph position="0"> In many NLP applications such as IR, we need to determine relations between terms. In most previous studies, one tries to determine the related terms to one single term (word). This makes the resulting relations ambiguous.</Paragraph>
    <Paragraph position="1"> Although several approaches have been proposed to remove afterwards some of the inappropriate terms, this only affects part of the noise, and much still remains. In this paper, we argue that the solution to this problem lies in the addition of context information in the relations between terms. We proposed to add another word in the condition of the relations so as to help constrain the context of application. Our experiments confirm that this addition of limited context information can indeed improve the quality of term relations and query expansion in IR.</Paragraph>
    <Paragraph position="2"> In this paper, we only compared biterm relations and unigram relations, the general method can be extended to triterm relations or more complex relations, provided that they can be extracted efficiently.</Paragraph>
    <Paragraph position="3"> This paper only investigated the utilization of context-dependent relations in IR. These relations can be applied in many other tasks, such as machine translation, word sense disambiguation / discrimination, and so on. These are some interesting research work in the future.</Paragraph>
  </Section>
class="xml-element"></Paper>