Mining high utility contrast patterns in sequences

Contrast sequential pattern mining (CSPM) is a critical field in data mining that concentrates on discovering contrasting information between different categories. However, existing CSPM methodologies often rely solely on frequency as the mining metric, which might not accurately mirror users’ authentic interests. In this study, we introduce a new framework that merges CSPM with the utility concept to extract utility-driven contrast patterns that can capture users’ genuine interests. We also discover an issue with pattern flipping, which can impact the differentiating ability of contrast patterns. We introduce two innovative utility-driven contrast pattern types according to the issue: Reduced stable contrast pattern (RSCP) and Flipping utility contrast pattern (FUCP). Furthermore, we present efficient mining algorithms, RCPMiner and FCPMiner, incorporating a new utility upper bound (MRSU) and novel pruning strategies. The experiments result shows our proposed algorithms to be effective and efficient and provide validation for the representational ability of RSCP. Improved performance is observed when RSCP is employed as the feature in the classification task, surpassing the performance achieved by utilizing other types of patterns. RSCP has also been demonstrated to possess lower redundancy when compared by the quantity of mined patterns.