Open Compliance Summit

Most license scanners are rule-based: Defined occurrences of text patterns let software identify a particular licensing in open source software. This approach is straight forward and easy to understand.

But there are two major drawbacks: first a license expert is required to define such rules. Secondly, since many licenses share common text portions imprecise license classifications occur.

In the FOSSology project, recently two new approaches
have been implemented: One approach is called Atarashi and uses information retrieval statistics about license texts and attempts to identify licensing statements based on computed statistics. The other approach named Rigel uses machine learning to understand licensing in source code, based on previously made clearing decisions.

This presentation covers the challenges of license recognition and explains how different approaches cover them.