API recommendation system in Software Engineering

Abstract

Software development depends on Application Programming Interfaces (APIs) to achieve their goals. However, choosing the right APIs remains as a difficult task for Software Engineers. In software engineering, recommendation systems are emerging to support Software Engineers in their decision-making tasks. Therefore, in this work, we proposed a methodology that considers software categories and recommends APIs to Software Engineers with software in initial (not using APIs) or advanced (using some APIs) stage of software development. We used collaborative filtering technique along with frequent Itemset mining technique for generating the corresponding large and top-N lists of APIs recommended. In the top-N lists, the goal was to find a few specific APIs that are supposed to be most useful to Software Engineers. In order to automate the methodology proposed, we developed a plug-in for the Eclipse IDE. In addition, we tested the methodology considering categories from the SourceForge open source repository. For every category, we evaluated large and top-N lists performance based on two classification accuracy metrics (precision and recall) and one efficacy metric (recall rate), obtaining promising outcomes. Thus, the results of evaluation metrics showed that our methodology could make useful API recommendations for Software Engineers with software that used a small number of APIs or did not use any API. Besides, our methodology was able to put relevant APIs even in high-ranking positions, even in small top-N lists of APIs recommended.