E-commerce platforms face challenges like fragmented user behavior and inefficient marketing resource allocation in new media. A multi-source data fusion architecture covering transactions, social platforms, and ads was built. Data from 237,000 users over 180 days was analyzed, identifying 5 user groups via K-Means and predicting conversion with 83.7% accuracy using XGBoost. Precision targeting boosted conversion by 37.3%, user lifetime value by 28.3%, ad ROI from 1:2.8 to 1:4.5, and reduced customer acquisition costs by 32.5%.