Towards scalable action rule discovery: a structured vertical partitioning method

Abstract Purpose: The extraction of actionable insights is critical for intelligent systems and recommendation engines. However, traditional methods for action rule discovery face challenges in scalability and efficiency when applied to large datasets. This study introduces a correlation-based vertical partitioning method to improve the consistency and interpretability of action rules while addressing the limitations of random partitioning and unstructured approaches. Methods: The proposed method clusters flexible attributes using correlations, enabling structured partitions for parallel rule generation via hierarchical clustering. Comparative experiments evaluated its precision, runtime, lightness, and coverage against random and baseline partitioning approaches. Results: The correlation-based method outperformed random partitioning and significantly improved runtime efficiency over the baseline. It generates interpretable rules in a single iteration, avoiding variability and repeated runs, though challenges in rule combination efficiency suggest areas for improvement. Conclusion: The correlation-based vertical partitioning method strikes a balance between computational efficiency and rule quality, making it a promising solution for large-scale action rule discovery. Future work could enhance scalability further by improving the rule combination process and exploring hybrid or adaptive partitioning strategies to extend the method’s applicability across diverse domains.