Web service recommendation for mashup creation based on graph network

In recent years, the world has witnessed the increased maturity of service-oriented computing. The mashup, as one of the typical service-based applications, aggregates contents from more than one source into a single user interface. Facing the rapid growth of the number of web services, choosing appropriate web services for dif-ferent mashup sources plays an important issue in mashup development, when, in particular, the new mashup is developed from the scratch. To solve this cold start problem when creating new mashups, we propose a web Service Recommenda-tion approach for Mashup creation based on Graph network, called SRMG. SRMG makes service recommendation based on service characteristics and historical usage. It first leverages Bidirectional Encoder Representations from Transformers, to intel-ligently discover mashups with similar functionalities based on specifications. After-ward, it employs GraphGAN to obtain representation vectors for mashups and ser-vices based on historical usage, and further obtains mashup preferences for each service based on representation vectors. Finally, the new mashup's preference for target services is derived from the preference of existing mashups that are similar to it. The extensive experiments on real datasets from ProgrammableWeb demonstrate that SRMG is superior to the state-of-the-art ones.