Mining actionable repetitive positive and negative sequential patterns

Repetitive positive and negative sequential patterns (PNSPs) recognize the repetitive characteristics of positive (occurring) and negative (nonoccurring) sequential patterns, thereby providing more comprehensive information than traditional PNSPs. However, existing repetitive PNSP mining methods produce numerous conflict patterns that do not benefit decision-making. To address this issue, we propose an actionable repetitive PNSP mining method, namely ARPNSP, for transaction databases. First, we propose the definition of negative occurrence under the self-adaptive gap and nonoverlapping conditions, which makes it possible to identify whether a pattern is actionable via correlation analysis. Second, we propose an offset sequence definition by adding a dummy character at the head of the sequences, which determines the population of repetitive PNSP. Finally, we utilize the bitmap structure to represent databases and occurrences of patterns, which avoids rescanning data sequences to calculate support. To the best of our knowledge, this study is the first attempt at actionable repetitive PNSP mining. Extensive experiments on real-world datasets show that ARPNSP can efficiently discover more actionable PNSPs with high correlation than the considered methods.