Fuzzy Similarity K-Type Prototype Algorithm and Marketing Methods

Abstract

In the field of user feature segmentation, the currently adopted segmentation methods have the defect of low segmentation accuracy. To address this problem, the study introduces the K-prototypes algorithm for user feature segmentation to improve the segmentation accuracy of user feature segmentation. The study first improves the traditional K-prototypes algorithm using fuzzy similarity matrix. The improved K-prototypes algorithm can effectively select the initial clustering center and fuzzy coefficients and weight coefficients, and pre-set the number of clusters in order to realize the accurate segmentation of user feature. After that, user feature segmentation model is constructed based on the improved K-prototypes algorithm to plan the best marketing methods for users with different characteristics. The study selected 605, 3200, and 684 data objects from the R15, D13, and credit approval datasets as experimental subject. Moreover, it compared the improved K-prototypes algorithm with the fuzzy C-means clustering algorithm and the density peak clustering algorithm in terms of clustering accuracy, root mean square error, mean absolute error, and clustering recall rate to evaluate the performance of the three algorithms. The performance advantages and disadvantages of the three algorithms were evaluated by accuracy, root mean square error, mean absolute error, and recall. The accuracy of the improved K-prototypes algorithm reached 0.9438, which was significantly higher than the other two algorithms. Moreover, the mean square error and mean absolute error of this algorithm were significantly lower than the other two algorithms, indicating that the clustering effect of this algorithm was significantly better than the other two algorithms. The recall of the improved K-prototypes algorithm reached 0.953, and the variation of recall was small, indicating the efficiency of this algorithm in dividing user features. All three algorithms were able to select the correct initial clustering center point for the improved K-prototypes algorithm under different dataset conditions, and the clustering purity of this algorithm was always maintained in the interval of 0.81-0.84. The outcomes reveal that the improved K-prototypes algorithm is able to accurately classify different users according to their characteristic requirements and can plan the best marketing methods for them.