Towards a Basic Principle for Ranking Effectiveness Prediction without Human Assessments: A Preliminary Study