Comparison of Feature Learning Methods for Metadata Extraction from PDF Scholarly Documents