<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1650">
  <Title>Automatically Assessing Review Helpfulness</Title>
  <Section position="11" start_page="429" end_page="429" type="ackno">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> Ranking reviews according to user helpfulness is an important problem for many online sites such as Amazon.com and Ebay.com. To date, most websites measure helpfulness by having users manually assess how helpful each review is to them. In this paper, we proposed an algorithm for automatically assessing helpfulness and ranking reviews according to it. Exploiting the multitude of user-rated reviews on Amazon.com, we trained an SVM regression system to learn a helpfulness function and then applied it to rank unlabeled reviews. Our best system achieved Spearman correlation coefficient scores of 0.656 and 0.604 against a gold standard for MP3 players and digital cameras.</Paragraph>
    <Paragraph position="1"> We also performed a detailed analysis of different features to study the importance of several feature classes in capturing helpfulness. We found that the most useful features were the length of the review, its unigrams, and its product rating. Semantic features like mentions of product features and sentiment words seemed to be subsumed by the simple unigram features. Structural features (other than length) and syntactic features had no significant impact.</Paragraph>
    <Paragraph position="2"> It is our hope through this work to shed some light onto what people find helpful in user-supplied reviews and, by automatically ranking them, to ultimately enhance user experience.</Paragraph>
  </Section>
class="xml-element"></Paper>