An energy-based comparative analysis of common approaches to text classification in the Legal domain