A framework for classifying and comparing source code recommendation systems

Abstract

The use of Application Programming Interfaces (APIs) is pervasive in software systems; it makes the development of new software much easier, but remembering large APIs with sophisticated usage protocol is arduous for software developers. Code recommendation systems alleviate this burden by providing developers with a ranked list of API usages that are estimated to be most useful to their development tasks. The promise of these systems has motivated researchers to invest considerable effort to develop many of them over the past decade, yet the achievements are not evident. To assess the state of the art in code recommendation, we propose a framework for classifying and comparing these systems. We hope the framework will help the community to conduct a systematic study to gain insight into how much code recommendation has so far achieved, in both research and practice.