Improving and Assessing the Fidelity of Large Language Models Alignment to Online Communities