Towards Interpretable and Efficient Automatic Reference-Based Summarization Evaluation