Feature Extraction Of Customer Behavior And Profit Mining Based On E-Business Model

With the development of International technology, E-Business systems bring more and more information for business and customers. To address this issue, a variety of personalized service were proposed and great attentions have been paid on this new technology, which has become a hotspot in recent researchers. From view of consumers, they care more about how to find the products which they need. An E-Business system could assist consumers to finish the process of purchase. From view of enterprises, it is important that finding high value products portfolios, optimization consumers groups, and creating more profit. In this thesis, we explored and researched some key issues of personalized service E-Business system. At the beginning of the thesis, main methods in data mining are presented and hot research in related field is reviewed. The basic theory and recent developments of evolutionary computation are also introduced. The rest chapters are the main contents and contributions of this thesis including:(1) A novel kind of intelligent E-Business model based personalized service is presented via systematic analyses and value chain thinking. It is then characterized and evaluated strategically in comparison with conventional models, which illustrates the advantages of this model in the E-Business environment for enterprises. Finally, this model is investigated through successful enterprises in E-Business, and a suite of policies for implementing this model are formulated to achieve strategical advantages.(2) A novel feature extraction methodology based purchase behavior is proposed. On first stage, we use Tanimoto similarity to measure purchase behavior similarity between customers, and design a clustering methodology based genetic algorithm to cluster customers who have similar purchase behavior to the same subpopulation. On the second stage, we design multi-population extraction customer feature methodology based genetic algorithm to find knowledge from all kind of subpopulation. In order to promote coevolution within the population and improve quality of rule set, we use q-nearest neighbor replacement policy and local search. In empirical study, we validate our methodology by using real-world data and compare with Apriori algorithm. It is demonstrated that our methodology could get a high quality rule set without generating large itemsets and rule is more flexible. At last, we analyze experiment results in detail.(3)We make use of association rule, and propose a multi-objective optimization model. The model considered both direct profit of product portfolio and cross selling profit. We design multi-objective genetic algorithm for finding global optimal solutions. In order to promote population diversity and local search ability, we design individual remedying and fill-up strategies and the local search technology. Then, we use real world data set to validate the proposed model and algorithm. From experimental results, the proposed model and algorithm could provide sufficient information for decision maker to design appropriate marketing policy.