Fixing Mislabeling by Human Annotators Leveraging Conflict Resolution and Prior Knowledge