<?xml version="1.0" standalone="yes"?>
<Paper uid="E06-1014">
  <Title>Improving Probabilistic Latent Semantic Analysis with Principal Component Analysis</Title>
  <Section position="8" start_page="111" end_page="111" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have presented LSA-PLSA, an approach for improving the performance of PLSA by leveraging the best features of PLSA and LSA. Our approach uses LSA to initialize a PLSA model, allowing for arbitrary weighting schemes to be incorporated into a PLSA model while leveraging the optimization used to improve the estimate of the PLSA parameters. We have evaluated the proposed framework on two tasks: personalsize information retrieval and text segmentation. The LSA-PLSAmodel outperformed PLSA on all tasks. And in all cases, combining PLSA-based models outperformed a single model.</Paragraph>
    <Paragraph position="1"> The best performance was obtained with combined models when one of the models was the LSA-PLSA model. When combining multiple PLSA models, the use of LSA-PLSA in combination with either two PLSA models or one PLSA and one LSA model improved performance while reducing the running time over the combination of four or more PLSA models as used by others.</Paragraph>
    <Paragraph position="2"> Future areas of investigation include quantifying the expected performance of the LSA-initialized PLSA model by comparing performance to that of the empirically best performing modeland examining whether tempered EMcould further improve performance.</Paragraph>
  </Section>
class="xml-element"></Paper>