A tree-based framework to mine top-K closed sequential patterns

Top-K closed sequential pattern (CSP) mining addresses the challenge of reducing the number of mined patterns and the dependency on the support threshold parameter. This study tackles top-K CSP mining from three angles: top-K generic CSPs, group CSPs, and redundancy-aware CSPs. We propose the novel SP-Tree-based KCloTreeMiner to mine these variations and introduce the PaMHep data structure for efficient candidate pattern maintenance. Two pruning strategies—namely, pattern absorption and SP-Tree-based temporary node projection—are also presented to reduce search space. This study offers a thorough theoretical analysis and establishes bounds for the top-K framework, covering everything from solution design to completeness and optimization. Evaluations on six real-life datasets show up to a 23% average runtime improvement for KCloTreeMiner over the benchmark algorithm TKCS. We also propose two greedy algorithms \(Max_{WC}\) $Ma{x}_{\mathit{WC}}$ and \(Max_{WOC}\) $Ma{x}_{\mathit{WOC}}$ for pattern summarization and introduce Subset Distance for measuring distances between sequential patterns, improving K-medoid clustering results over average silhouette-width for the reported clusters.