Research on the Application of Optimized Clustering and Feature Fusion Techniques in E-commerce Customer Segmentation and Retention Strategies

In the field of e-commerce, achieving precision marketing is the key to improving core competitiveness, and customer segmentation is the premise for achieving precision marketing and formulating retention strategies. Clustering algorithms and feature fusion technology can achieve customer segmentation. However, how to design an algorithm that not only performs customer segmentation but also enables adaptive optimization of its parameters remains a challenge. In this paper, a kernel k-means clustering algorithm based on deep Q network (DQN-KKM) is proposed. First, the recency, frequency and monetary (RFM) model is used to extract three-dimensional features of e-commerce customers: the interval since their last consumption, consumption frequency, and consumption amount. Secondly, the kernel k-means clustering algorithm is defined to process nonlinear separable data, which can map the data to the implicit space and achieve clustering after completing the similarity calculation through the kernel function. Finally, we propose the problem of adaptive optimization of the number of clusters and kernel function parameter, and establish a Markov decision process (MDP) to analyze the issue. The deep Q network (DQN) is trained to obtain the optimal number of clusters and kernel function parameters to achieve the optimization of the clustering algorithm. Experimental validation confirmed the algorithm’s performance, and effective retention strategies were devised based on the experimental results.