Model-based Preference Optimization in Abstractive Summarization without Human Feedback