<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1302">
  <Title>Adaptive String Similarity Metrics for Biomedical Reference Resolution</Title>
  <Section position="8" start_page="14" end_page="15" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have explored a set of string similarity metrics in the biological domain in the service of reference resolution. String similarity is only one parameter to be considered in this task. We presented encouraging results for assigning strings to UMLS concepts based solely on string similarity metrics -- demonstrating that adaptive string similarity metrics show significant promise for biomedical text processing.</Paragraph>
    <Paragraph position="1"> Further progress will require a system that 1) utilizes context of occurrence of respective strings for handling ambiguity and 2) further improves recall 6Inspection of the data indicates that the purely character-based methods are more robust than one might think. There are at least 8 strings to match against for a concept and it is likely that at least one of them will have similar word order to the test string.</Paragraph>
    <Paragraph position="2">  through expanded synonyms.</Paragraph>
    <Paragraph position="3"> Future work should also consider the dependent nature (via transitivity) of reference resolution. Comparing a test string against all (current) members of an equivalence class and considering multiple, similar test instances simultaneously (McCallum and Wellner, 2003) are two directions to pursue in this vein.</Paragraph>
  </Section>
class="xml-element"></Paper>