Generalizability of Code Clone Detection on CodeBERT

TL;DR

This paper investigates the generalizability of CodeBERT for code clone detection, revealing significant performance drops when applied to different code snippets and functionality IDs beyond the training data.

Contribution

It provides an empirical evaluation of CodeBERT's generalizability across diverse Java code clones, highlighting limitations in semantic clone detection.

Findings

Significant F1 score drop on different code snippets

Reduced performance on unseen functionality IDs

Challenges in semantic clone detection with CodeBERT

Abstract

Transformer networks such as CodeBERT already achieve outstanding results for code clone detection in benchmark datasets, so one could assume that this task has already been solved. However, code clone detection is not a trivial task. Semantic code clones, in particular, are challenging to detect. We show that the generalizability of CodeBERT decreases by evaluating two different subsets of Java code clones from BigCloneBench. We observe a significant drop in F1 score when we evaluate different code snippets and functionality IDs than those used for model building.