Hierarchical RNN Networks for Structured Semantic Web API Model Learning and Extraction

RESTful Web APIs have no description files like WSDL in traditional Web service. Although some REST API definition models have been arising recently, there is still lacking in structured description format for existing large mounts of Web APIs. Almost all Web APIs are documented in semi-structured web pages, and these documentation formats are various for different sites. It's hard for machine to read the semantics of Web APIs. In this paper, we have proposed a novel hierarchical recurrent neural network to convert REST API documentation to structured machine-readable description format -- the Swagger REST API specification. The network extracts the Swagger defined attributes of a REST API from HTML web pages without any feature engineering. With the extracted API specifications, we built an API repository to index, search and compose Web APIs. Experiment showed that the hierarchical RNN model performed well even with only a few training samples.