College event ecosystems have characteristics such as being volatile, short-lived, and being greatly molded by the influence of peers. This research work presents a unified, student-focused recommendation model that is hinged on Cross-View Semantic Contrastive Learning with Graph Neural Networks (CSCL-GNN). The model simultaneously exploits a user–event interaction graph and a user–user social graph, learns scalable embeddings with LightGCN, and uses a multi-head attention module for weighting that part of the influential neighbors. A new, cross-view semantic fusion step which helps to find the commonality of prototypes (clusters) between interaction and social views thus considerably eases the cold-start problem for new students or events. The training here combines a pairwise ranking objective (BPR) and graph-level contrastive losses (InfoNCE) that really tighten the tie between the structural and the attention-refined representations. The performed test using the real college event participation logs demonstrates that the proposed model shows considerable improvements of the top-K recommendation metrics (HR@K, NDCG@K) and particularly high gains in the cold-start scenarios (up to \(\sim \) 14% HR@10 relative to baselines).