<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1112">
  <Title>AnyQ: Answer Set based Information Retrieval System</Title>
  <Section position="5" start_page="6" end_page="6" type="concl">
    <SectionTitle>
6. Conclusion
</SectionTitle>
    <Paragraph position="0"> The accuracy of IR result continues to grow on importance as exponential growth of WWW, and it is therefore increasingly important that appropriate retrieval technologies be developed for the web. We have introduced a new type of IR, Answer Set based IR, attempts to provide high quality answer documents to user queries.</Paragraph>
    <Paragraph position="1"> In the context of answer set-driven text retrieval, it is crucial to capture semantics and pragmatics of sentences in user queries and documents. In our case, we defined the semantic category of the answer as attributes, the documents associated with each attributes as answer set. We attempted to provide more accurate answers by attaching attributes to individual concepts in concept network. In order to construct knowledge bases, a certain level of quality is guaranteed, we developed a new method for attributed-based classifier(ABC) and built attribute pattern for improving accuracy of ABC and query processing both. In retrieval, we process a natural language query, extract concepts and attributes, and map them to the knowledge base so that the answer documents associated with the &lt;concept, attribute&gt; pairs can be retrieve.</Paragraph>
    <Paragraph position="2"> Our proposed IR ranked highly relevant document on top result, thus it helps reducing human efforts dramatically to find answer. By established operational system, named AnyQ, our experiment showed realistic possibility of our approach systematically.</Paragraph>
    <Paragraph position="3"> While our experiments were designed carefully, and comparisons made thoroughly, it has limitations. Our current work depends on the domain of the concept network. It is not clear how the proposed method can be extended to other domains. Our assumption, reflecting semantics in sentence to &lt;concept, attribute&gt; pairs, needs to be tested further. More fundamentally, we need a certain amount of manual work to initially construct the knowledge base such as the concept hierarchy and the initial training documents. We will have to see how the initial manual process influences the latter processes and what kind of performance degradation occurs when smaller efforts are used for the initial construction.</Paragraph>
  </Section>
class="xml-element"></Paper>