In response to the problems of rough user segmentation, inaccurate churn prediction, low recommendation efficiency, and low conversion rate in traditional new media marketing, this paper proposes a new media user profile construction and precision marketing strategy based on the C-ATT-LSTM model. Firstly, construct a complete new media user profile from three dimensions: basic attributes, behavioral attributes, and statistical attributes; Secondly, design a C-ATT-LSTM network that integrates convolution, attention mechanism, and LSTM to achieve high-dimensional feature extraction and temporal dependency capture; Thirdly, SOM clustering is used to stratify users and independently model and identify high churn users for each group; Finally, by combining collaborative filtering and content recommendation, personalized coupons can be accurately pushed. The experiment was validated using real data from an online daily necessities store from July 2024 to July 2025.7, and the results showed that the proposed method significantly outperformed baseline algorithms such as LR-RF, ANN, and Adaboost in accuracy and F1 score; The ablation experiment showed that the complete solution achieved a user conversion rate of 13.69%, a conversion rate growth rate of 387.66%, an estimated sales increase of 5.68% compared to the benchmark, and the optimal input-output ratio. Research has shown that the proposed model can accurately characterize user characteristics, efficiently identify churn risks, and provide feasible technical solutions for precision marketing of new media platforms
