<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3808">
  <Title>Seeing stars when there aren't many stars: Graph-based semi-supervised learning for sentiment categorization</Title>
  <Section position="7" start_page="51" end_page="51" type="evalu">
    <SectionTitle>
5 Discussion
</SectionTitle>
    <Paragraph position="0"> We have demonstrated the benefit of using unlabeled data for rating inference. There are several directions to improve the work: 1. We will investigate better document representations and similarity measures based on parsing and other linguistic knowledge, as well as reviews' sentiment patterns. For example, several positive sentences followed by a few concluding negative sentences could indicate an overall negative review, as observed in prior work (Pang and Lee, 2005). 2. Our method is transductive: new reviews must be added to the graph before they can be classified. We will extend it to the inductive learning setting based on (Sindhwani et al., 2005). 3. We plan to experiment with cross-reviewer and cross-domain analysis, such as using a model learned on movie reviews to help classify product reviews.</Paragraph>
  </Section>
class="xml-element"></Paper>