Open Vocabulary Learning on Source Code with a Graph-Structured Cache

TL;DR

This paper introduces a graph-structured cache to handle open vocabulary challenges in source code modeling, significantly improving performance on code completion and variable naming tasks using graph neural networks.

Contribution

It proposes a novel graph-structured cache mechanism that enhances neural models' ability to handle open vocabulary in source code, improving accuracy in code understanding tasks.

Findings

Over 100% improvement in variable naming accuracy

Enhanced performance on code completion tasks

Moderate increase in computation time

Abstract

Machine learning models that take computer program source code as input typically use Natural Language Processing (NLP) techniques. However, a major challenge is that code is written using an open, rapidly changing vocabulary due to, e.g., the coinage of new variable and method names. Reasoning over such a vocabulary is not something for which most NLP methods are designed. We introduce a Graph-Structured Cache to address this problem; this cache contains a node for each new word the model encounters with edges connecting each word to its occurrences in the code. We find that combining this graph-structured cache strategy with recent Graph-Neural-Network-based models for supervised learning on code improves the models' performance on a code completion task and a variable naming task --- with over $100\%$ relative improvement on the latter --- at the cost of a moderate increase in computation time.