<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3809">
  <Title>Random-Walk Term Weighting for Improved Text Classification</Title>
  <Section position="7" start_page="58" end_page="59" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> Based on results obtained in text classification experiments, the TextRank random-walk model to term weighting was found to achieve error rate reductions of 3.5-44% as compared to the traditional frequency-based approach. The evaluation results have shown that the system performance varies depending on window size, dataset, as well as classifier, with the greatest boost in performance recorded for KNN ,Rocchio, and SVM. We believe that these results support our claim that random-walk models can accurately estimate term weights, and can be used as a technique to model the probabilistic distribution of features in a document.</Paragraph>
    <Paragraph position="1"> The evaluations reported in this paper has shown that the TextRank model can accurately provide uni-gram probabilities for a sequence of words. In future work we will try to extend the TextRank model and use it to define a formal language model in which we can estimate the probability of entire sequences of words (n-grams).</Paragraph>
  </Section>
class="xml-element"></Paper>