Features
• Refined
–  Lexical: keyword lists
for specific fields.
–  Position: first 15 lines
of the article (simulates
the header of article)
–  Email: handwritten
regular expressions
–  Numeric: both digit and
word based
–  Orthographic:
capitalization feature
• Baseline
–  Vocabulary: lexical
tokens +/- window of 2
Purposefully simple to assess the
efficacy of CRFs and features