This paper divides the research modes of consumer purchase behavior characteristics into three categories: experience-driven mode, theory-driven mode, and data-driven mode. An analysis algorithm based on customer consumption behavior is proposed, and the idea of combining customer consumption behavior factors such as satisfaction and loyalty is proposed. Through comparison, it is pointed out that the data-driven model is most suitable for analyzing the characteristics of online consumers’ purchasing behavior. Using the decision support of knowledge base, different service schemes for customers with different evaluation degrees are realized. In order to improve the accuracy of sample classification and maximize the output function, genetic algorithm is used to optimize the samples. A deep neural network structure algorithm is proposed to classify customer transaction data samples. In this algorithm, the sheep nodes are not fixed, but the number of hidden layers and unit nodes of the neural network are dynamically determined according to the sample training. The research excavates various kinds of valuable information such as consumer preferences and consumption structure from the huge consumption data of consumers. It is not only helpful for enterprises to analyze consumers’ consumption behavior and organize production but also helpful for enterprises to realize the concept of personalization.