I am planning to use graph-based techniques for the task of fault localization in code.
The training dataset consists of buggy code snippets as well as their corresponding fixed code snippets. The testing dataset contains only the buggy code.The goal is to identify the line at which the bug might be occurring. Does anyone know how to go about using GCN for this task? What should the node embeddings be? Thank you!
I recommend this paper as a starter: [1711.00740] Learning to Represent Programs with Graphs. You might also find more advanced approaches following this line.
See also Graph Neural Networks in Program Analysis from the same author.
As an irrelevant note, LLMs look like a viable alternative to GNNs as well. See [2301.08653] An Analysis of the Automatic Bug Fixing Performance of ChatGPT (note that this paper is uploaded before GPT-4 is released) for instance.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.