<p>With the rapid development of recommendation systems, the need for personalized training content has grown, yet existing systems often struggle with data sparsity and cold-start issues, limiting their ability to deliver robust recommendations. In view of this challenge, we propose a novel approach that integrates deepfake-assisted generation model into recommendation systems to enhance personalized training recommendations. By generating more available data through similarity-based deepfake data generation techniques, the system expands the scope of training scenarios, improving recommendation robustness and addressing cold-start challenges. Later, we leverage collaborative filtering and content-based filtering to tailor training content based on the integration of real data and generated data. At last, comparison evaluation is made based on real-world datasets and the final experimental results demonstrate the system’s capability to provide high-quality, personalized training recommendations while enhancing overall user experience especially in the sparse-data environment.</p>