Objective To investigate relations among different aspects in supervised word sense disambiguation (WSD; supervised machine learning for disambiguating the sense of a term in a context) and compare supervised WSD in the biomedical domain with that in the general English domain.

Methods The study involves three datasets (a biomedical abbreviation data set, a general biomedical term data set, and a general English data set). We implemented four machine learning algorithms including i) Naive Bayes (NBL) and Decision Lists (TDLL), ii) our adaptation of Decision Lists (ODLL), and iii) our Mixed Supervised Learning (MSL). There were six feature representations (i.e., various combinations of collocations, bag of words, oriented bag of words, etc.) and five window sizes (2,4,6,8,10).

Results Supervised WSD is suitable only when we have enough sense-tagged instances with at least a few dozens of instances for each sense. The combination of collocations and neighboring words are appropriate selections for the context. For terms with biomedical unrelated senses, a large window size such as the whole paragraph should be used, while for general English words a moderate window size between 4 to 10 should be used. The performance of our implementation of decision list classifiers for abbreviations was better than that of traditional decision list classifiers. However, the opposite held for the other two sets. Also, our mixed supervised learning was stable and generally better than others for all sets.

Conclusion From this study, we found that different aspects of supervised WSD depend on each other. The experiment method presented in the study can be used to select the best supervised WSD classifier for each ambiguous term.