Smart campus management data mining and personalized service push strategy based on machine learning empowering spark

This study constructs a multi-source data mining system for smart campus, and based on Spark distributed computing, forms three core models: user portrait, demand prediction and personalized push. The clustering results show the hierarchical characteristics of behavior structure, and the Silhouette coefficient reaches 0.71, indicating that there are clear differences among user groups. The demand forecasting model obtains F1 values of 0.85 and 0.80 in the academic early warning and resource demand tasks, respectively, which reflects the behavior trend with high identifiability. The CTR and CVR of collaborative filtering model in practical experiments increased from 9.5 to 18.3 and 2.1 to 4.7%, which showed the effectiveness of personalized push in content matching. The throughput of Spark cluster in the 32-node environment reaches 83,000 records per second, and the average delay drops to 0.94 s, which provides a stable operation foundation for real-time analysis. The research forms a deployable data analysis framework to provide a structured technical path for teaching support, resource allocation and campus governance.