In plain words: A tool turns JavaScript code into a web of linked pieces and uses a graph neural network, which follows those links, to guess each piece's type for automatic code repair. It got over 90% of types right, beating earlier attempts at this task.

Abstract · Inferring Javascript types using Graph Neural Networks

The recent use of `Big Code' with state-of-the-art deep learning methods offers promising avenues to ease program source code writing and correction. As a first step towards automatic code repair, we implemented a graph neural network model that predicts token types for Javascript programs. The predictions achieve an accuracy above $90\%$, which improves on previous similar work.

Jessica Schrouff, Kai Wohlfahrt, Bruno Marnette, Liam Atkinson
arXiv:1905.06707 · cs.LG, cs.PL, cs.SE, stat.ML · submitted May 16, 2019
abstract · pdf · html · Published at the Representation Learning on Graphs and Manifolds ICLR 2019 workshop (https://rlgm.github.io/papers/)