Analyzing Source Code using Graph Neural Networks and Natural Language

While computers are becoming an integral part of our lives, programming them still remains a highly specialized skill. The last few years there is an increased research interest in methods that focus on the intersection of programming and natural language processing (NLP), that aim to help create machine learning-based tools that aid software engineers by understanding source code’s natural language components and allow end-users to employ natural language to interact with computers.

Within this research area, Graph Neural Networks (GNN) have shown promising results in exploiting the rich structure and long-range dependencies in source code. In this talk, I will discuss three machine learning architectures that employ GNNs for source code-related tasks including code summarization, code generation and change representation. Then, I will illustrate how these networks can find applications in NLP tasks, such as natural language summarization. Finally, I will conclude with a discussion of some of the open challenges on source code-related tasks and research on graph neural networks.