<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2915">
  <Title>Which Side are You on? Identifying Perspectives at the Document and Sentence Levels</Title>
  <Section position="8" start_page="114" end_page="114" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this paper we study a new problem of learning to identify the perspective from which a text is written at the document and sentence levels. We show that much of a document's perspective is expressed in word usage, and statistical learning algorithms such as SVM and na&amp;quot;ive Bayes models can successfully uncover the word patterns that reflect author perspective with high accuracy. In addition, we develop a novel statistical model to estimate how strongly a sentence conveys perspective, in the absence of sentence-level annotations. By introducing latent variables and sharing parameters, the Latent Sentence Perspective Model is shown to capture well how perspectives are reflected at the document and sentence levels. The small but positive improvement due to sentence-level modeling in LSPM is encouraging. In the future, we plan to investigate how consistently LSPM sentence-level predictions are with human annotations.</Paragraph>
  </Section>
class="xml-element"></Paper>