Digital Libraries, Joint Conference on (2003)
Houston, Texas USA
May 27, 2003 to May 31, 2003
Atsuhiro Takasu , National Institute of Informatics
In this paper, we propose a method for extracting bibliographic attributes from reference strings captured using Optical Character Recognition (OCR) and an extended hidden Markov model. Bibliographic attribute extraction can be used in two ways. One is reference parsing in which attribute values are extracted from OCR-processed references for bibliographic matching. The other is reference alignment in which attribute values are aligned to the bibliographic record to enrich the vocabulary of the bibliographic database. In this paper, we first propose a statistical model for attribute extraction that represents both the syntactical structure of references and OCR error patterns. Then, we perform experiments using bibliographic references obtained from scanned images of papers in journals and transactions and show that useful attribute values are extracted from OCR-processed references. We also show that the proposed model has advantages in reducing the cost of preparing training data, a critical problem in rule-based systems.
Atsuhiro Takasu, "Bibliographic Attribute Extraction from Erroneous References Based on a Statistical Model", Digital Libraries, Joint Conference on, vol. 00, no. , pp. 49, 2003, doi:10.1109/JCDL.2003.1204843