Revealing the Unseen: Guiding Personalized Diffusion Models to Expose Training Data