How many preprints have actually been printed and why: a case study of computer science preprints on arXiv