The proliferation of Web APIs has made selecting suitable ones for Mashup development increasingly difficult. Web API recommendation systems can help developers quickly discover the desirable Web APIs. A wide variety of methods have been proposed, and among them, graph neural network (GNN)-based approaches have gained prominence. Despite their success, these methods often suffer from noise accumulation and over-smoothing due to repeated convolutional aggregations, and their performance is further hindered by the inherent sparsity of interaction data, leading to biased node representations. To address these limitations, we propose a novel textual similarity-supervised graph collaborative filtering model for Web API recommendation, named TSSGCF. TSSGCF leverages textual descriptions to construct a refined, denoised graph, mitigating noise propagation and alleviating over-smoothing during graph convolution. Furthermore, TSSGCF introduces an innovative textual similarity-supervised learning mechanism, which enhances node representation learning, significantly improving recommendation accuracy. Experimental validation on a real-world dataset demonstrates the effectiveness of the proposed model. To facilitate the reproducibility of our work, we have open-sourced our code at https://github.com/IntelligentServiceLab/TSSGCF .