Understanding Political Polarisation using Language Models: A dataset and method