EVALUATION
To test our approach, we developed a proof-of-concept implementation.
We evaluated it on data from 10 APIs from the IBM WatsonTM Developer Cloud.
The APIs expose a set of natural language processing and machine learning capabilities such as visual recognition, language identification, or machine translation.
1 Our test data comprises logs of actual usage of the APIs, as well as Swagger 2.0 specifications manually written by the API developers, and which we use as the ground truth.
We also compared our results to those obtained using a competing open-source solution. In the rest of the paper, we refer to the APIs as #1 to #10.