Comparing representations of long clinical texts for the task of patient note-identification