The rapid growth of the mobile application ecosystem, driven by the widespread adoption of mobile internet technology, has led to a highly competitive market with millions of apps and third-party libraries. Developers face challenges in efficiently identifying suitable libraries while ensuring product quality. Existing research ignores the utilization of the topological information implicit in attribute information and relies solely on bipartite interaction graphs to introduce self-supervised signals. To address this limitation, this paper proposes CALibRec, a graph contrastive learning method with category awareness for third-party library recommendation. CALibRec constructs homogeneous graphs based on co-occurrence frequencies then generates augmented views for the homogeneous graphs and integrates category information as heterogeneous nodes into homogeneous graphs. By joint training with the recommendation and contrastive learning tasks, the model improves recommendation accuracy. Experimental results demonstrate that CALibRec improves the accuracy of third-party library recommendations.