Self-supervised Preference Optimization: Enhance Your Language Model with Preference Degree Awareness