Mining Interesting Negative Sequential Patterns Based on Influence

Abstract

Negative sequential pattern (NSP) mining can capture frequently occurring and non-occurring behavior information and can play an irreplaceable role in many applications. Most traditional NSP mining algorithms adopt a support measure to discover interesting patterns. However, the support measure does not truly reflect the interestingness of patterns in some cases. In particular, it ignores the effect of the support of every element and the order characteristics among these elements. Hence, an influence measure was proposed to truly reflect the interestingness of patterns. However, the current influence measure is used only in positive sequential pattern (PSP) mining and does not involve NSPs. To address these problems, this study proposes an algorithm, InfI-NSP, to mine interesting NSPs based on influence. First, we modify an existing NSP mining algorithm to efficiently mine NSPs. Second, we modify the influence measure and apply it to NSP mining to mine interesting NSPs. To the best of our knowledge, InfI-NSP is the first algorithm to mine interesting NSPs based on influence. Experiments on real-life and synthetic datasets show that InfI-NSP is effective.