<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1069">
  <Title>E ective Structural Inference for Large XML Documents</Title>
  <Section position="8" start_page="14" end_page="14" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have addressed the problem of structural inference for large XML documents. In doing so we began by motivating the research and reviewing the literature. The use of Minimum Message Length as a measure for the quality of inferred content models has been introduced, adapted from work in related elds.</Paragraph>
    <Paragraph position="1"> This measure has proven to be an appropriate and a vast improvement over previous subjective techniques. We have also presented the rst wide spread comparison of di erent grammatical inference techniques. This involved the implementation of the existing Alergia, k-contextual, (k, h)-contextual and sk-strings methods, as well as the creation of new algorithms. The new methods include the Greedy strategy, the ACO meta-heuristic, Stochastic Hill Climbing and our proposed, hybrid sk-ANT heuristic. Comprehensive experimental data revealed that our proposed method was the most e ective and most stable of the methods, followed by the sk-strings heuristic. From this work we may conclude that the problem of structural inference is both important and tractable. For current applications we recommend use of the sk-ANT technique. The use of MML as a quality measure is also recommended, due to its generality and objectivity.</Paragraph>
  </Section>
class="xml-element"></Paper>