The Case for API Communicability Evaluation: Introducing API-SI with Examples from Keras

Abstract

In addition to their vital role in professional software development, Application Programming Interfaces (APIs) are now increasingly used by non-professional programmers, including end users, scientists and experts from other domains. Therefore, good APIs must meet old and new user requirements. Most of the re-search on API evaluation and design derives from user-centered, cognitive perspectives on human-computer interaction. As an alternative, we present a lower-threshold variant of a previously proposed semiotic API evaluation tool. We illustrate the procedures and power of this variant, called API Signification Inspection (API-SI), with Keras, a Deep Learning API. The illustration also shows how the method can complement and fertilize API usability studies. Additionally, API-SI is packaged as an introductory semiotic tool that API designers and researchers can use to evaluate the communication of design intent and product rationale to other programmers through implicit and explicit signs thereof, encountered in the API structure, behavior and documentation.