Predicting user purchase behavior using shopping history data on e-commerce platforms helps to improve user experience and marketing effect. Our paper uses the time-sliding window method to construct features that mine users’ interest preferences in different periods based on the real interaction records between users and products in e-commerce scenarios. Then, a model for predicting user purchase behavior based on CNN-LSTM is proposed. By automatically extracting and selecting user attributes, product attributes, and user behavioral features, the model is used to predict user purchasing behavior. An online retail platform implements precision marketing using this model. The results show that the calculated values of the marketing effect in the Attention Stage, Interest Stage and Active Participation Stage are between [0.8-1.0], and the effect of Precision Marketing is “Excellent”. The calculated value of the marketing effect in the action stage and repeat purchase stage is between [0.6-0.8], and the effect of precision marketing is “good”. After the implementation of precision marketing, the operating income of e-commerce platform A is increasing, while the operating expense ratio remains stable. This paper’s model can effectively improve consumers’ purchase intention, as evidenced by its findings.