Perspectives of machine learning and natural language processing on characterizing positive energy districts