<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0116">
  <Title>Chinese Named Entity Recognition with Conditional Random Fields</Title>
  <Section position="7" start_page="119" end_page="120" type="evalu">
    <SectionTitle>
5 Evaluation Results
5.1 Results on Sighan bakeoff 2006
</SectionTitle>
    <Paragraph position="0"> We evaluated our system in the close track, on all three corpora, namely Microsoft Research (MSRA), City University of Hong Kong (CityU), and Linguistic Data Consortium (LDC). Our official Bakeoff results are shown at Table 3, where the columns P, R, and FB1 show precision, recall and F measure(b = 1). We used all three types of features in our final system.</Paragraph>
    <Paragraph position="1"> In order to evaluate the contribution of features, we conducted the experiments of each type of features using the test sets with gold-standard dataset. Table 4 shows the experimental results,  where F1 refers to use basic features, F2 refers to use the word boundary features, F3 refers to use the char features, and Post refers to perform the post-processing.</Paragraph>
    <Paragraph position="2"> The results indicated that word boundary features helped on LDC and CityU, char features only helped on CityU and the post-processing always helped to improve the performance.</Paragraph>
  </Section>
class="xml-element"></Paper>