High accuracy citation extraction and named entity recognition for a heterogeneous corpus of academic papers

Citation indices are increasingly being used not only as navigational tools for researchers, but also as the basis for measurement of academic performance and research impact. This means that the reliability of tools used to extract citations and construct such indices is becoming more critical; however, existing approaches to citation extraction still fall short of the high accuracy required if critical assessments are to be based on them. In this paper, we present techniques for high accuracy extraction of citations from academic papers, designed for applicability across a broad range of disciplines and document styles. We integrate citation extraction, reference parsing, and author named entity recognition to significantly improve performance in citation extraction, and demonstrate this performance on a cross-disciplinary heterogeneous corpus. Applying our algorithm to previously unseen documents, we demonstrate high F-measure performance of 0.98 for author named entity recognition and 0.97 for citation extraction.