Automatic metadata generation: a comparison of two annotators
Jacquelynn K. Sherman · Carolina Digital Repository (University of North Carolina at Chapel Hill) · 2019
There is a growing need to develop effective techniques and tools for automatic metadata generation. The research presented in this master's paper compares the annotation functions of NCBO BioPortal with those of HIVE in order to determine whether basic term matching techniques or machine learning techniques produce higher quality results. The research was conducted by selecting a document set and testing it on both annotators. The metadata generated by the annotators was then assessed by three human evaluators in terms of relevance, precision, specificity, and exhaustivity. The research found that on average the results produced by the HIVE annotator, which employed machine learning techniques, had 10 percent higher specificity, 17 percent higher exhaustivity, and 19.4 percent higher precision than the results produced by BioPortal. The paper concludes that the machine learning underlying HIVE produces higher quality results than basic term matching techniques, and that this approach deserves greater research attention.