<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1107">
  <Title>Feature Selection in Categorizing Procedural Expressions</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Related Works
</SectionTitle>
    <Paragraph position="0"> The questions related in all procedures were addressed by an expert system(Barr et al., 1989). However, in QA and information retrieval for open domain documents from the Web, the system requires a more flexible and more machine-operable approach because of the diversity and changeable nature of the information resources. Many competitions, e.g.</Paragraph>
    <Paragraph position="1"> TREC and NTCIR, are being held each year and various studies have been presented (Eguchi et al., 2003; Voorhees, 2001). Recently, the most successful approach has been to combine many shallow clues in the texts and occasionally in other linguistic resources. In this approach, the performance of passage retrieval and categorization is vital for the performance of the entire system. In particular, the productiveness of the knowledge of expressions corresponding to each question type, which is principally exploited in retrieval and categorization, is important. In this perspective, that means that the requirements for categorization in such applications are different from those in previous categorizations.</Paragraph>
    <Paragraph position="2"> Many studies have been made that are related to QA.</Paragraph>
    <Paragraph position="3"> Fujii et al.(2001) studied QA and knowledge acquisition for definition type questions. Approaches by seeking any answer text in the pages of FAQs or newsgroups appeared in some studies(Hamada et al., 2002; Lai et al., 2002). Automatic QA systems in a support center of organizations was addressed in a study by Kurohashi et al.(2000).</Paragraph>
    <Paragraph position="4"> However, most of the previous studies targeting QA address fact type or definition type questions, such as &amp;quot;When was Mozart born?&amp;quot; or &amp;quot;What is platinum?&amp;quot;. Previous research addressing the type of QA relevant to procedures in Japanese is inconclu- null sive. In text categorization research, the feature selection has been discussed(Taira and Haruno, 2000; Yang and Pedersen, 1997). However, most of the research addressed categorization into taxonomy related to domain and genre. The features that are used are primarily content words, such as nouns, verbs, and adjectives. Function words and frequent formative elements were usually eliminated. However, some particular areas of text categorization, for example, authorship identification, suggested a feasibility of text categorization with functional expressions on a different axis of document topics.</Paragraph>
    <Paragraph position="5"> From the perspective of seeking methods of domain-independent categorization for QA, this paper investigates the feasibility of functional expressions as a feature for the extraction of lists including procedural expressions.</Paragraph>
  </Section>
class="xml-element"></Paper>