Utilizing domain-based features to improve classification accuracy of biomedical text having bacterial associations

Biomedical text can serve as a crucial source for obtaining a deeper understanding of the complexities pertaining to inter-bacterial associations. Existing methods are either based on bacterial co-occurrence observed in texts or rely on generic text mining methods, both of which may lead to extraction of limited amount of information. In this article, the efficiency of a multitude of generic text mining approaches for classifying text having bacterial associations have been first evaluated on a manually curated corpus. Further, a set of domain (or field of microbiology) specific features accompanied by three dictionaries relevant to bacterial associations were introduced. Incorporation of these domain-specific features in addition to generic text mining features from abstracts as well as sentences were shown to significantly improve the classification accuracy.