With the rapid proliferation of Web services and Web APIs, recommendation systems can effectively address the issue of information overload and alleviate the burden of meaningless filtering. Existing approaches can help filtering appropriate Web services for mashup creation, however, they often fall short of developers’ different and personalized needs by recommending only a fixed number of APIs and lack precision in aligning mashup requirements across all categories. To solve the above issue, this paper introduces a novel Web service recommendation framework called AWAR for mashup creation, which focuses on the matching strategy between mashup requirements and Web APIs, and enhances recommendation effectiveness by integrating natural language processing, optimization algorithms, and deep learning. Extensive experiments conducted on large-scale real datasets demonstrate that the proposed approach receives superior recommendation results on multiple evaluation metrics compared to advanced competing baselines.