<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1016">
  <Title>Convolution Kernels with Feature Selection for Natural Language Processing Tasks</Title>
  <Section position="10" start_page="0" end_page="0" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> This paper proposed a statistical feature selection method for convolution kernels. Our approach can select significant features automatically based on a statistical significance test. Our proposed method can be embedded in the DP based kernel calculation process for convolution kernels by using sub-structure mining algorithms.</Paragraph>
    <Paragraph position="1">  Experiments show that our method is superior to conventional methods. Moreover, the results indicate that complex features exist and can be effective. Our method can employ them without over-fitting problems, which yields benefits in terms of concept and performance.</Paragraph>
  </Section>
class="xml-element"></Paper>