A Hybrid Preference Learning and Context RefinementArchitecture