Thesaurus anomaly detection by user action monitoring
Bitencourt, J.L.; Cancian, P.ín.S.; Pacheco, E.J.; Nohama, P.; Schulz, S.
Studies in Health Technology and Informatics 129(Part 1): 655-659
2007
ISSN/ISBN: 0926-9630 PMID: 17911798 Document Number: 606184
The construction and maintenance of a medical thesaurus is a non-trivial task, due to the inherent complexity of a proper medical terminology. We present a methodology for transaction-based anomaly detection in the process of thesaurus maintenance. Our experiences are based on lexicographic work with the MorphoSaurus lexicons, which are the basis for a mono- and cross-lingual biomedical information retrieval system. Any "edit"or "delete" actions within these lexicons that undo an action defined earlier were defined as anomalous. We identify four types of such anomalies. We also analyzed to which extent the anomalous lexicon entries had been detected by an alternative, corpus-based approach.