Pattern set mining entails discovering groups of frequent itemsets that represent potentially relevant knowledge. Global constraints are commonly enforced to focus the analysis on most interesting pattern sets. However, these constraints evaluate and select each pattern set individually based on its itemset characteristics. This paper extends traditional global constraints by proposing a novel constraint, called schema-based constraint, tailored to relational data. When coping with relational data itemsets consist of sets of items belonging to distinct data attributes, which constitute the itemset schema. The schema-based constraint allows us to effectively combine all the itemsets that are semantically correlated with each other into a unique pattern set, while filtering out those pattern sets covering a mixture of different data facets or giving a partial view of a single facet. Specifically, it selects all the pattern sets that are (i) composed only of frequent itemsets with the same schema and (ii) characterized by maximal size among those corresponding to that schema. Since existing approaches are unable to select one representative pattern set per schema in a single extraction, we propose a new Apriori-based algorithm to efficiently mine pattern sets satisfying the schema-based constraint. The experimental results achieved on both real and synthetic datasets demonstrate the efficiency and effectiveness of our approach.