<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1093">
  <Title>Summarizing Encyclopedic Term Descriptions on the Web</Title>
  <Section position="6" start_page="5" end_page="5" type="evalu">
    <SectionTitle>
5 Discussion
</SectionTitle>
    <Paragraph position="0"> The goal of our research is to automatically compile a high-quality large encyclopedic corpus using the Web. Hand-crafted encyclopedias lack new terms and new definitions for existing terms, and thus the quantity problem is crucial. The Web contains unreliable and unorganized information and thus the quality problem is crucial. We intend to alleviate both problems. To the best of our knowledge, no attempt has been made to intend similar purposes.</Paragraph>
    <Paragraph position="1"> Our research is related to question answering (QA). For example, in TREC QA track, definition questions are intended to provide a user with the definition of a target item or person (Voorhees, 2003). However, while the expected answer for a TREC question is short definition sentences as in a dictionary, we intend to produce an encyclopedic text describing a target term from multiple viewpoints.</Paragraph>
    <Paragraph position="2"> The summarization method proposed in this paper is related to multi-document summarization (MDS) (Mani, 2001; Radev and McKeown, 1998; Schiffman et al., 2001). The novelty of our research is that we applied MDS to producing a condensed term description from unorganized Web pages, while existing MDS methods used newspaper articles to produce an outline of an event and a biography of a specific person. We also proposed the concept of viewpoint for MDS purposes.</Paragraph>
    <Paragraph position="3"> While we targeted Japanese technical terms in the computer domain, our method can also be applied to other types of terms in different languages, without modifying the model. However, a set of viewpoints and patterns typically used to describe each view-point need to be modified or replaced depending the application. Given annotated data, such as those used in our experiments, machine learning methods can potentially be used to produce a set of viewpoints and patterns for a specific application.</Paragraph>
  </Section>
class="xml-element"></Paper>