Automated Filtering of Human Feedback Data for Aligning Text-to-Image Diffusion Models