E-commerce Intelligent Recommendation Optimization and Personalized Marketing Strategy Based on Big Model

This paper proposes an e-commerce recommendation model integrating large-scale representation learning with personalized marketing strategies. Leveraging multimodal fusion of text, image, and category data, semantic alignment, and dynamic ranking, the framework incorporates optimized fusion selection, task-specific weighting, and reinforcement learning – based marketing decision-making. Multi-task learning jointly models clicks, conversions, and other behaviors, with weight tuning verified through comparative experiments. A cold-start evaluation is included to assess adaptability for new users and items. Experimental results show CTR improving from 6.45% to 9.17%, CVR from 1.89% to 3.25%, and NDCG@10 reaching 0.832, confirming enhanced accuracy, robustness, and user value. CCS CONCEPTS Applied computing ~ Electronic commerce ~ commerce infrastructure