Utility Multidimensional Association Rule Mining Based On Customer Segmentation

With the increasingly competition,how to better maintain the existing customers and develop more potential customers as much as possible,is the problem faced by every enterprise decision-makers.Subdividing the customer according to the existing customer purchase records,and mining customer purchase behavior with the help of multidimensional association rule,helps decision makers to precise marketing strategy.Therefore,customer segmentation,multidimensional association rules mining technology has attracted more and more attention.In this paper,we combined customer segmentation and multi-dimensional association rule mining to analyze the behaviors of different customer.First,we subdivided the customers based on the existing data.Then we made each classification of customers as a dimension,and combined it with other attributes to construct the multi-dimensional data model.Finally,we used the multi-dimensional association rule mining technology to mine purchase behavior of different classification of customers.We combined the RFM model and K-means clustering model,subdivided the customer according to the existing customer purchase records.and the three attributes of recency,frequency and monetary are used as the customer value evaluation indicators.We use local outlier factor instead of traditional method to classify the recency,frequency and monetary indicators.In this way,not only the data can be preserved more completely,but also the clustering results can be more accurate.Finally,through measuring the degree of similarity inner different classification and the degree of difference between different classification with two indicators,compactness and dispersion,we can find the best clustering results and apply it to the multidimensional association rule mining.The traditional multidimensional association rule mining determines the validity of rules by the rule's frequency,and measured by the support and confidence.This mining method only considers the statistical correlation betweenrules and ignores the semantic importance which is the effectiveness that the rules can bring.In this paper,we introduce the utility function as a comprehensive measure of statistical correlation and semantic significance.The utility function mainly measures the effectiveness of the rule from three aspects: opportunity,probability and effectiveness.Probability represents the statistical correlation,opportunity and effectiveness represents the semantic significance.The results show that the rules mined by the utility function not only meet the objective requirements of higher frequency,but also have the subjective expectations of higher effectiveness.