Towards Unified Modeling for Positive and Negative Preferences in Sign-Aware Recommendation