<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1641">
  <Title>Sentiment Retrieval using Generative Models</Title>
  <Section position="4" start_page="345" end_page="345" type="intro">
    <SectionTitle>
6 concludes the paper.
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Some efforts for the TREC Novelty Track were related to our work. Although some of the topics used in the Novelty Track in 2003 and 2004 (Soboroff and Harman, 2003; Soboroff, 2004) were related to opinions, most of the efforts were focused on topic, such as studies using term distribution within each sentence, e.g., (Allan et al., 2003; Losada, 2005; Murdock and Croft, 2005).</Paragraph>
    <Paragraph position="1"> Amongst the participants in the TREC Novelty Track, only (Kim et al., 2004) proposed a method specialized to opinion-bearing sentence retrieval, by making use of lists of words with positive or negative polarities. They aimed to find opinions on a given topic but did not distinguish or did not care about sentiment polarities that should be represented in some sentences (hereafter, opinion retrieval). We focus on finding positive views or negative views according to a given topic and sentiment of interest (hereafter, sentiment retrieval). Our work is the first work on sentiment retrieval, to the best of our knowledge.</Paragraph>
    <Paragraph position="2"> In the context of sentiment classification, some researchers have conducted studies on the topic dependence of sentiment polarities. (Nasukawa and Yi, 2003) and (Yi et al., 2003) extracted positive or negative expressions on a given product name using handmade lexicons. (Engstr&amp;quot;om, 2004) studied how the topic dependence influences the accuracy of sentiment classification and attempted to reduce the influence to improve the accuracy.</Paragraph>
    <Paragraph position="3"> (Wilson et al., 2005) investigated how context influences sentiment polarity at the phrase level in a corpus, beginning with a predefined list of words with polarities. Their focus on the phenomena of topic dependence of sentiment can be shared with our work; however, their work is not directly related to ours, because we focus on a different task, sentiment retrieval, where different approaches are required.</Paragraph>
  </Section>
class="xml-element"></Paper>