Re-ranking Using Large Language Models for Mitigating Exposure to Harmful Content on Social Media Platforms