Mining Readme Files to Support Automatic Building of Java Projects in Software Repositories

Abstract

Automatic building of software projects provides a desirable foundation to support a large variety of software engineering research tasks based on open software repositories. In this paper, we propose the first technique to automatically extract software build commands from software readme files and Wiki pages, and combine the extracted commands for software building. Specifically, we leverage the Named Entity Recognition (NER) technique for build-command extraction, and prioritize the extracted build commands to identify which one should be used in software build. Our experiment on top Java projects from GitHub reveals that, the proposed technique can correctly identify more than 90% of build commands, and can successfully build 84% of the projects that can be built successfully through manual inspection of software support documents.