Measuring quality of similarity functions in approximate data matching