Abstract: "Web Application Programming Interface(API) allows third-party and subscribed users to access data and functions of a software application through the network or the Internet. Web APIs expose data and functions to the public users, authorized users or enterprise users. Web API providers publish API documentations to help users to understand how to interact with web-based API services, and how to use the APIs in their integration systems. The exponential raise of the number of public web service APIs may cause a challenge for software engineers to choose an efficient API. The challenge may become more complicated when web APIs updated regularly by API providers.In this paper, we introduce a novel transformation-based approach which crawls the web to collect web API documentations (unstructured documents). It generates a web API Language model from API documentations, employs different machine learning algorithms to extract information and produces a structured web API specification that compliant to Open API Specification (OAS) format. The proposed approach improves information extraction patterns and learns the variety of structured and terminologies. In our experiment, we collect a sheer number of web API documentations. Our evaluation shows that the proposed approach find RESTful API documentations with 75\% accuracy, constructs API endpoints with 84\%, constructs endpoint attributes with 95\%, and assigns endpoints to attributes with an accuracy 98\%. The proposed approach were able to produces more than 2,311 OAS web API Specifications."
Authors: Mehdi Bahrami and Wei-Peng Chen (Fujitsu Laboratory of America Inc., USA)
Email: email@example.com, firstname.lastname@example.org