英语论文网

cision tree rule sets at time t,tþkrti; rtþkj each rule from the corresponding ruleset Rt, Rtþk, where i¼1; 2; . . . ; jRtj; j ¼1; 2; . . . ; Rtþk supt(ri) support of ri in time t data setDong and Li (1999) introduced the emergingpatterns concept that captures significant changes anddifferences between data sets. The emerging patterns aredefined as item sets whose supports increase significantlyfrom one data set to another. More specifically, theemerging patterns are item sets whose growth rates arelarger than a given threshold value. When applied to timestampeddatabases, the emerging patterns can captureemerging trends in business or demographic data.We bringfrom the study of Dong and Li (1999) the term emergingpattern with the following modified definition for ourresearch.Definition 1: Emerging patterns For rule rtþkj , if thefollowing two conditions are met, then we call it the ruleof emerging patterns with respect to rti:(1) Conditional and consequent parts are the samebetween rti; rtþkj .(2) Supports of the two rules are significantly different.Example 1rti: If Income¼High, Age¼High, then Sales amount¼High (support¼0.1)rtþkj : If Income¼High, Age¼High, then Sales amount¼High (support¼0.13)In this case, rtþkj is the emerging pattern with respect to rtiifwe specify the minimum growth rate to be 0.2. This isbecause the two rules have the same rule structure and theirgrowth rate is 0.3.The other type of change is unexpectedness, which isfound from many studies about discovering interestingpatterns (Liu & Hsu, 1996; Silberschatz & Tuzhilin, 1996;Liu et al., 1997; Suzuki, 1997; Padmanabhan & Tuzhilin,1999). The unexpected changes can be found from newlydiscovered rules, which are different from users’ existingbeliefs. Liu and Hsu (1996) defined the unexpected changesas rule similarity and difference aspects. The approach isbased on a syntactic comparison between a rule and abelief. In their study, a rule and a belief are ‘different’ if theconsequents are ‘far apart’ but the conditions are ‘similar’,where ‘similarity’ and ‘difference’ are defined syntacticallybased exclusively on the structure of the rules. We redefineunexpected changes adapting from the study by Liu andHsu (1996).Definition 2: Unexpected changes (or unexpected consequentchanges) rtþkj is unexpected change with respect to rtiif the conditional parts of rti; rtþkj are similar but theconsequent parts of the two rules are quite different.Example 2rti: If Income¼High, Age¼High, then Sales amount¼Highrtþkj : If Income¼High, Age¼High, then Sales amount¼LowIn this case, rtþkj is an unexpected consequent change withrespect to rtisince the conditional parts of rti; rtþkj are similarbut the consequent parts of the two rules are quite different.Other types of change are the added rules and theperished rules (Lanquillon, 1999). An added rule is a newlyarisen rule which could not be found in the past and aperished rule is a disappeared rule, which can only be foundin the past but not the present. The added and perishedrules are defined as follows.Definition 3: Added rules=Perished rules rtþkj is anadded rule if all the conditions and consequents are quitedifferent from any of rtiin Rt, and rtiis a perished rule if allthe conditions and consequents are quite