<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1116">
  <Title>Extraction of User Preferences from a Few Positive Documents</Title>
  <Section position="6" start_page="222" end_page="222" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this study, we apply fuzzy inference technique and term reweighting scheme based on the term co-occurrence similarity to the problem that extract important keywords representing contents of documents presented by users. We have conducted extensive experiments on the Reuters-21578 collection. The results show that our method outperforms two well-known training algorithms for linear text classifiers. Moreover, some variants of our method have been explored to analyze the characteristics of our method. Though this paper only describes how to extract user preferences from example documents, the technique will be applicable to several areas such as query modification in IR, user profile modification in information filtering, text summarization and so forth directly or with some modifications.</Paragraph>
    <Paragraph position="1"> Since only positive examples are considered in our method, the method is not applicable to a document set containing negative examples. For covering negative examples, it needs to modify the fuzzy inference rules with considering additional input variables. The proposed method was also designed for a small set of documents. So, we could not achieve performance improvement as described in this paper when our method is applied to a large set of documents. However, such a problem will be alleviated if clustering techniques are used together as in (Alberto et al., 2001; Lam and Ho, 1998; Ugur et al., 2000).</Paragraph>
  </Section>
class="xml-element"></Paper>