To Analyze and Regulate Human-in-the-loop Learning for Congestion Games