Existing Web API search engines allow for only category-based browsing and keyword or tag-based searches for RESTful services without offering the capability of discovering and composing real-world RESTful services from the viewpoint of application developers. Therefore, we propose a novel approach, referred to as TAD (Transformation-Annotation-Discovery), to address the above issue. TAD firstly transforms OpenAPI (Swagger) documents of RESTful services into the graph structure in graph database, and provides an annotation engine to automatically annotate the semantic concepts on each graph node by using LDA (Latent Dirichlet Allocation), and WordNet. Next, TAD conducts service composition based on the user requirement by the two modules, service discovery chain and pipeline-based composition. The service discovery chain checks service interface compatibility and retrieves supported and aided services to bridge the gap between the user requirement and the current discovered services based on the Hungarian algorithm. The pipeline-based composition module finds services that semantically fit the user's required tasks based on the annotated graph database and sends candidate services to service discovery chains to simultaneously seek for multiple possible composition solutions fitting the user's composition requirement. Experimental results show that the proposed approach is with good performance under the precision metric.