Mining Quantitative Top-k Sequential Patterns under Constraints

Sequential pattern mining with quantitative attributes is essential in many real-world applications where items are associated with numerical values such as quantities, dosages, or utilities. While numerous algorithms have been proposed for frequent or top-k sequential pattern mining, none simultaneously address top-k extraction, quantitative constraints, and flexible user-defined constraints within a unified framework. In this paper, we introduce a novel Constraint Programming (CP) model for mining top-k quantitative sequential patterns under heterogeneous constraints. Our approach relies on a global constraint that enforces both subsequence embedding and frequency filtering through dynamic support raising. We propose two dedicated search strategies that accelerate convergence toward the k most frequent patterns while progressively pruning the search space. Extensive experiments on real-world datasets demonstrate that our approach scales favorably compared to state-of-the-art top-k sequential pattern mining algorithms in terms of scalability and flexibility, while naturally supporting complex structural and quantitative constraints.