Reproduce “Deep Code Search with Naming-Agnostic Contrastive Multi-View Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiadong Feng, Wei Li, Suhuang Wu, Zhao Wei, Yong Xu, Juhong Wang, Hui Li
Researchers at Xiamen University and Peking University developed NACS, a naming-agnostic deep code search framework that uses contrastive multi-view learning on Abstract Syntax Trees (ASTs) to overcome performance degradation due to variable naming variations. NACS consistently achieved higher MRR scores across datasets, demonstrating robustness to name changes while maintaining comparable efficiency to other pre-training methods.
Key insights
- Learning code representations directly from AST structural properties, rather than variable names, significantly improves robustness against naming variations.
- Combining graph-view and path-view representations of ASTs via contrastive multi-view learning enriches the understanding of intrinsic code structure.
- Customized semantic and syntactic data augmentation strategies for ASTs are critical for generating diverse yet semantically equivalent views for self-supervised learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.