Improving textual medication extraction using combined conditional random fields and rule-based systems