Springer retracts nearly 40 publications on 'bonkers' dataset