<?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="5" start_page="557" end_page="558" type="relat">
    <SectionTitle>
5. Related Work
</SectionTitle>
    <Paragraph position="0"> Co-occurrence analysis is a common method to determine term relations. The previous studies have been limited to relations between two words, which we called unigram relations. This expansion approach has been integrated both in traditional retrieval models (Jing and Croft, 1994) and in LM (Berger and Lafferty 1999). As we observed, this type of relation will introduce much noise into the query, leading to unstable effectiveness.</Paragraph>
    <Paragraph position="1"> Several other studies tried to filter out noise expansion (or translation) terms by considering the relations between them (Gao et al., 2002;  Jang et al. 1999; Qiu and Frei, 1993; Bai et al. 2005). However, this is insufficient to detect all the noise. The key issue is the ambiguity of relations due to the lack of context information in the relations. In this paper, we proposed a method to add some context information into relations. (Lin, 1997) also tries to solve word ambiguity by adding syntactic dependency as context.</Paragraph>
    <Paragraph position="2"> However, our approach does not require determining syntactic dependency. The principle of our approach is more similar to (Yarowsky, 1995). Compared to this latter, our approach is less demanding: we do not need to identify manually the exact word senses and seed context words. The process is fully automatic. This simplification is made possible due to the requirement for IR: only in-context related words are required, but not the exact senses.</Paragraph>
    <Paragraph position="3"> Our work is also related to (Smadja and McKeown, 1996), which tries to determine the translation of collocations. Term combinations or biterms we used can be viewed as collocations.</Paragraph>
    <Paragraph position="4"> Again, there is much less constraint for our related terms than translations in (Smadja and McKeown, 1996).</Paragraph>
  </Section>
class="xml-element"></Paper>