Self-Supervised Source Code Annotation from Related Research Papers

Pierre Haritz, Lukas Pfahler, Thomas Liebig, Helena Kotthaus

Abstract: Language analysis of scientific documents and analysis of source code have been done independently in the past. This work presents a network architecture and a self-supervised training approach to find alignments between published computer science research papers and their corresponding public source code by learning a representation of encodings from transformers, from which source code can be enriched with helpful information. We present our ideas, findings and plans for upcoming research.