In order to accurately identify and predict consumers’ purchasing intentions, preferences, and behavior patterns, this paper uses CNN technology to explore consumer behavior characteristics in the live e-commerce environment and build a prediction model. Through data collection and preprocessing, this paper collects behavioral data such as user viewing time, interaction frequency, purchase records, etc. on the live platform, and cleans and extracts features from the data. Then, the CNN model is used to learn the features of user behavior data. CNN extracts features such as the spatiotemporal correlation and behavior intensity of user behavior patterns in the data, and then inputs the extracted features into the fully connected layer of the model to classify and predict behavior patterns, and output possible user purchase behaviors and preference trends. In addition, this paper also introduces a real-time monitoring mechanism to train and adjust the model online by dynamically updating user behavior data. In terms of behavior prediction accuracy, the accuracy of the CNN model is much higher than that of the SVM. In the MSE comparison, the average MSE value of CNN is about 315.7, which is much lower than the 894.1 of the SVM. In the MAE comparison, the highest MAE value of CNN is 50 and the lowest is 10, while the MAE value of SVM is between 56–120. These results show that the CNN model has higher accuracy and lower error in predicting consumer behavior, and can more accurately predict consumer consumption behavior.