In a manner of keyword search, a software developer could develop mashup applications with more sophisticated functions by choosing a set of Web APIs from an extensive pool of available options, which can surprisingly save the development costs to ensure up-to-date mashup development. However, when faced with a large and high variety of Web APIs, developers often encounter several challenges, such as functional incompatibility and limited diversity. Moreover, although the number of Web APIs is enormous, available interaction datasets are extremely sparse, which may increase the risk of development failures. Recently, contrastive learning (CL) performs well in dealing with data sparsity problems by comparing the original and augmented representations learned from a bipartite graph. Thus, we propose a CL-based Diversity-Aware Web APIs Recommendation approach to recommend diversified and compatible Web APIs for mashup creation. Specifically, C-DAWAR first constructs a Web APIs correlation graph to identify the minimal group Steiner trees within the constructed graph. It then learns node representation using contrastive learning in a self-supervised manner. In particular, after the graph convolution operation in the contrastive learning pipeline, C-DAWAR applies the self-attention mechanism to better capture global features. Finally, Determinantal Point Processes (DPP) is employed to enhance the diversity of the recommended results. Comprehensive experimental results on widely used real-world datasets demonstrate the effectiveness of C-DAWAR.