Research On Techniques Of Repetitive Negative Sequential Pattern Mining With Gap Constraints

Repetitive sequential pattern mining considers the characteristic that patterns occur repeatedly between sequences and within a sequence,where the repetitive positive sequential pattern focuses on the occurring(positive)events,while the repetitive negative sequential pattern further focuses on the non-occurring(negative)events.Repetitive negative sequential patterns play an important role in many transaction services.For example,insurance fraud detection based on repetitive collision payments that are not repaired at a specified location.However,the current studies on repetitive negative sequential pattern mining are few and strictly constrained,and do not achieve good application results.In addition,these studies only consider the frequency of pattern occurrence,resulting in many conflicting patterns in the mining results that are not beneficial for decision-making.To solve the above problems,this thesis proposes related methods that enable mining repetitive negative sequential patterns from transaction sequence databases,avoiding generating conflicting sequential patterns and mining sequential rules considering the repetitive occurrence of patterns,taking repetitive negative sequential patterns with non-overlapping element occurrence positions and self-adaptive gap values between two consecutive elements as the research object.The main innovative contributions of this thesis are as follows:(1)To address the problem that the existing methods have stricter constraints and cannot obtain more useful information,a two-phase self-adaptive gap and non-overlapping repetitive negative sequential pattern mining method SN-RNSP is proposed.First,the method uses the bitmap structure to represent databases and maintain pattern candidates,which avoids scanning the database repeatedly for pattern support calculation.Second,a bitmap-based repetitive positive sequential pattern mining method BM-RPSP is proposed.Finally,based on the existing negative containment definition and the BM-RPSP method,a fast method SN-RNSP is proposed to mine repetitive negative sequential patterns using set theory and bitmap indexing.Experiment results on real-world and synthetic datasets show that SN-RNSP can effectively find more repetitive negative sequential patterns in transaction databases under non-overlapping and self-adaptive gap constraints.(2)To address the problem of conflicting patterns obtained by existing methods,an efficient two-phase non-overlapping self-adaptive gap actionable repetitive negative sequential pattern mining method ARNSP is proposed.First,the method proposes a definition of negative sequential pattern occurrence under non-overlapping and self-adaptive gap constraints and improve the existing definition of offset sequences to determine the population of sequential pattern occurrences,and then introduce the correlation coefficient to filter out those uncorrelated or weakly correlated sequential patterns.Second,an actionable repetitive positive sequential pattern mining method ARPSP is proposed.Finally,based on the ARPSP method,a method ARNSP is proposed to mine actionable repetitive negative sequential patterns by scanning the bitmap representation of databases.Experiment results show that ARNSP can discover more correlation-based and actionable negative sequential patterns in transaction databases.(3)To address the problem that existing methods only consider the occurrence frequency of patterns but not enough to make predictions and recommendations,a mining rule from non-overlapping self-adaptive gap repetitive positive and negative sequential pattern method RPnsp Rule is proposed.First,the method generates sequential rule candidates from the repetitive sequential patterns where all or any of the antecedents and consequents are negative sequential patterns.Second,to calculate the correlation between the antecedent and the consequent,the method proposes a definition of offset sequences oriented to the negative containment adapted by SN-RNSP to determine the population of sequential pattern occurrences.For each rule candidate,it only selects rules with positive correlation between antecedents and consequents.Finally,the method normalizes the confidence to satisfy the support-confidence framework.Experiment results show that RPnsp Rule can effectively discover more positive and negative sequential rules in transaction databases for decision making.