k-modes聚类方法的改进与在可追溯猪肉消费偏好研究中的应用

Improvement and Application of k-modes Clustering Method in Traceable Pork Consumption Preference Research

可追溯食品消费偏好等实证调查中具有大量的分类属性的数据,而常用的k-means聚类方法主要适用于连续的数值型数据,难以精确分析可追溯食品消费者类别.为此,引入k-modes聚类方法,通过改进聚类精度,改善聚类流程,改进了现有的k-modes算法,以更好地应用于分类属性数据的聚类分析.以分类正确率、类精度、召回率和平均迭代次数为评价检验的具体指标,相关检验表明,与经典k-modes聚类算法等相比较,改进后的k-modes聚类算法在标准数据集上实验结果良好.在此基础上,以可追溯猪肉为案例,通过菜单选择实验法收集消费者对可追溯猪肉信息属性偏好的数据,建立仿真分析流程,运用改进后的k-modes算法进行聚类分析,研究了消费者对可追溯猪肉属性的群体性偏好.仿真结果显示,消费者对可追溯猪肉信息属性的偏好具有明显的层次性、差异性,可基于消费偏好将消费者划分为4个类别.

There are large amounts of categorical data in empirical investigation of consumption preference for traceable food. However, the commonly used k-means clustering algorithm is mainly applied to continuous numeric data, it is difficult to accurately analyze consumer categories of traceable food. In this paper, the k-modes clustering method is introduced to better applied to cluster categorical data by improving clustering accuracy, clustering process, and modifying the current k-modes algorithm. The accuracy, precision, recall, and average number of iterations are used as specific indicators for evaluation. The tests show that the improved k-modes algorithm obtains better results on standard data set compared with the conventional clustering, based on which, the data of consumers’ preference for traceable pork were collected by menu-based choice experiment. The simulation analysis process was established, and the improved k-modes algorithm was used for clustering analysis to study the group preference of consumers for traceable pork. The simulation results show that there is an obvious hierarchy and heterogeneity in consumers’ preference for the traceability information attributes, and consumers can be divided into four classes based on the differences in consumers’ preference.