Leveraging Textual and Non-Textual Features for Documentation Decluttering

Abstract

This paper describes the participation of a team from the University of Bari in the Decluttering Challenge organized in the scope of the DocGen2 workshop. We propose a supervised approach relying on a minimal set of non-textual features (length, overlapping between the comment text and the source code, code block type, tags, comment type) and classical textual features (bag-of-words). Our system ranked 2nd in the documentation decluttering task.

Publication
2020 IEEE International Conference on Software Maintenance and Evolution