<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0503">
  <Title>The Problem of Precision in Restricted-Domain Question-Answering. Some Proposed Methods of Improvement</Title>
  <Section position="3" start_page="2" end_page="2" type="intro">
    <SectionTitle>
3 Methods for Improving Precision
</SectionTitle>
    <Paragraph position="0"> The first approach to improve the precision performance of the IR module is to use a better engine, e.g. by adjusting the parameters, modifying the formulas of the engine, or replacing a generic engine by a more domain-specific one, etc.</Paragraph>
    <Paragraph position="1"> Now suppose that the IR engine is already fixed, e.g. because we have achieved the best engine, or, more practically, because we cannot make changes or afford another engine. The second approach consists in improving the results returned by the IR engine. One main direction is candidate re-ranking, i.e. pushing good candidates in the returned candidate list to the first ranks as much as possible, thus increasing Q(n). To do this, we need some information that can characterize the relevance of a candidate to the corresponding question better than the IR engine did. The most prominent kind of such information may be the domain-specific language used in the working domain of the QA system, particularly its vocabulary, or even more narrowly, its terminological set.</Paragraph>
    <Paragraph position="2"> In the following, we will present our development of the second approach on the Bell Canada QA system first, because it seems less costly than the first one. However, we will present some implementations of the first approach later.</Paragraph>
  </Section>
class="xml-element"></Paper>