r, we developed a decision-tree-based methodologyto detect changes of customer behavior automaticallyfrom customer profiles and sales data at different timesnapshots. For this purpose, we found rule sets fromdecision trees and defined three types of change. Then wedeveloped the similarity and difference measures for rulematching to detect all types of change in syntactic aspects.Additionally, the degree of change was evaluated to detectsignificantly changed rules in semantic aspects.We summarize the opportunities of using this methodologyand various applications in practical business perspectivesas follows. First, with regard to macro aspects,business managers can follow the change trends using oursuggested change detection methodology. They need toanalyze their customers’ changing behavior in order toTable 8: Number of changed rules in the new decision treeat time t and time tþkType of change Number ofchangedrulesNumber of significantchanged rulesEmerging patterns 1 1 (degree of change >0.1)Unexpected changes 0 0 (degree of change >0.1)Added=Perished rules 9 9 (degree of change >0.001)Table 9: Significant emerging patterns (degree of change>0.1) in the new decision tree at time t and timetþksupt(ri) suptþk(rj) aij1 If Sales visit¼Low,Payment¼Cash,Reserved money¼High, thenSales amount¼Low0.24 0.11 0.5417Table 10: Significant added=perished rules (degree of change >0.001) in the new decision tree at time t and time tþkrtiMSV Support aij1 If Sales visit¼High, then Sales amount¼High 0 0.2965 0.29652 If Sales visit¼Low, Payment¼Card, then Sales amount¼Low 0 0.2269 0.22693 If Sales visit¼Low, Reserved money¼Low, then Sales amount¼Low 0 0.1997 0.19974 If Sales visit¼High, Age¼20s, then Sales amount¼High 0 0.1806 0.18065 If Sales visit¼High, Age¼40s, Order count¼High, then Sales amount¼High 0 0.0891 0.08916 If Sales visit¼Low, Reserved money¼Low, Payment¼Card, then Sales amount¼Low 0 0.0762 0.07627 If Sales visit¼Low, Payment¼Cash, Reserved money¼Low, then Sales amount¼Low 0 0.0694 0.06948 If Sales visit¼High, Age¼40s, Order count¼Low, then Sales amount¼High 0 0.0082 0.00829 If Sales visit¼High, Age¼Teen, then Sales amount¼Low 0 0.0034 0.0034Expert Systems, September 2005, Vol. 22, No. 4
provide products and services that suit the changing needsof the customers. Second, with regard to micro aspects, it ispossible for a manager to understand customer needs moredeeply and design additional niche-marketing campaignsbased on the rule sets of the suggested methodology.Knowing the purchasing history of a certain customersegment can give a better understanding of the behavior ofthe segment.Change detection is more suitable in domains where theenvironment is relatively dynamic and there is muchhuman intervention. Besides understanding customerbehavior change, another promising application is analyzingthe effectiveness of a marketing campaign. If a managergenerates rules from the sales data set before and after acampaign, he=she can evaluate the effectiveness of his=hermarketing campaign by comparing two rule sets using thesuggested methodology. Furthermore, the suggested methodologycan be used recursively in a time-series data set toanalyze the change of classified customer s
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。