Search

This article proposes an approach to multiple attribute decision making with partially known attribute weight information where individual assessments are provided as interval-valued intuitionistic fuzzy numbers (IVIFNs). By establishing an optimization model based on the deviation degree, the proposed approach derives a linear program for determining attribute weights. The weights are subsequently used to synthesize individual IVIFN assessments into an aggregated IVIFN value for each alternative. In order to rank alternatives based on their aggregated IVIFN values, a novel method is developed for comparing two IVIFNs by introducing the formula of possibility degree and the ranking vector of the possibility degree matrix. Finally, an illustrative outsourcing decision problem is employed to verify the proposed approach and to demonstrate its practicality and effectiveness.

In order to solve the classification problems of customer behaviors with randomicity and non-conformability, a combination classification method is proposed of multiple decision trees based on genetic algorithm. In this method, multiple decision trees that adopt the method of probability measurement level output are combined in parallel. Genetic algorithm is utilized to optimize connection weight matrix in combination algorithm. Furthermore, two sets of simulation experiment data are used to test and evaluate the proposed method. Results of the experiments indicate that the proposed method generates a higher classification accuracy rate than other methodspsila for customer behavior segmentation.

Nowadays, population features differ from place to place in China. The differences between population and customer samples, which are called "spatial population drift," are caused by economic, social and cultural reasons. Prediction model adaptability of customer segmentation is weakened by spatial population drift. A Bayesian network can effectively combine prior knowledge and sample data information. This paper proposes a method of a Bayesian network to improve model adaptability for different samples while keeping its high accuracy. It sets model structure without parameters as general form on population or large samples with good data quality, and then trains model parameters by local sample. By this way, the prediction model for special customer instance is built. It can solve the uncertainty of customer segmentation based on customer data in China to some extent.

The evaluation of e-commerce Web sites is the main measure of knowing the value, comparing the quality and guiding the construction of the Web sites. With the increasing use of e-commerce Web sites, it is gaining more and more attention, and becoming the focus of study. Of all the evaluation methods, analytic hierarchy process (AHP) is favored by many researchers because of its strict mathematical principles, and considering comprehensive features at the same time. But as we all know, in the process of using AHP we should set up a comparison matrix by comparing pairs of criteria or subcriteria. When there are many Web sites involved, it is very possible that the compare results disagree with each other. However, the two new methods, grey analysis (GA) and concordance analysis (CA) demonstrate their advantages for their demand of comparably few data, simple calculation, and straightaway principles. Furthermore the two methods are easy to achieve programming. This paper presents that in the process of evaluation, we can first apply the two methods to evaluate the e-commerce Web sites. If the results of the two evaluations are the same, we accept the results. Otherwise we use AHP to select the better one of the best results acquired by using the above two methods. By using the combination of the three methods, we acquire a more objective and believable evaluation result. In the fourth part of the paper, we prove that the result of this comprehensive method is the same as only using AHP by applying it to a case. And this method would be much easier when it achieves programming in the future

In the Internet shopping environment, changes of customer's needs grow increasingly outstanding. For discovering the changes, the paper mines the transaction databases of different time periods by using association rule discovery, and extracts the association rules and discovers the changes in network customer behavior by comparison and analysis between the two sets of association rules. This paper presents a new algorithm that contains the first support, the second support and Rel-confidence. The algorithm solves the problem of the present algorithms cannot discover association rules with infrequent data items. And according to the changes of network customer's behavior, robust pattern, appearing pattern and unexpected change are presented to measure the changes.

Two types of errors, 'rejecting true and accepting false' are inevitable in customer value segmentation. The traditional data mining method which is on the total accuracy rate cannot reflect the influence caused by the great difference of misclassification costs and unbalanced quantity distribution of customers who have various values. The thesis proposes two enhancements. The first is developing the cost-sensitive Support Vector Machine (SVM) classifier by presenting misclassification cost function based on customer value, which is evaluated by the function of the exceptional lost. The second enhancement is a method of multi-category classification based on the binary classification version of SVM. The data test result proves that the method can control the different types of errors distribution with various cost of misclassification accurately, reduce the total misclassification cost and distinguish the customer value effectively.