Extending inverse frequent itemsets mining to generate realistic datasets: complexity, accuracy and emerging applications