Interpreting longitudinal data through temporal abstractions: An application to diabetic patients monitoring