CSSG: Measuring Code Similarity with Semantic Graphs

2026-01-07

arXiv.org

AI Summary

Existing code similarity metrics are often confined to string-level or syntactic representations, failing to capture deep semantic relationships. This work proposes CSSG, a novel approach that, for the first time, incorporates program dependence graphs into code similarity modeling. By explicitly encoding control dependencies and variable interactions, CSSG constructs a semantics-aware representation of code. The method integrates control-flow analysis with graph representation learning and demonstrates significant improvements over state-of-the-art metrics on the CodeContests+ dataset. It achieves superior performance in discerning semantic similarity both within a single programming language and across different languages, offering a more accurate and robust measure of functional equivalence.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Cognitive Modeling & Cognitive Systems: Analogy
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing code similarity metrics, such as BLEU, CodeBLEU, and TSED, largely rely on surface-level string overlap or abstract syntax tree structures, and often fail to capture deeper semantic relationships between programs.We propose CSSG (Code Similarity using Semantic Graphs), a novel metric that leverages program dependence graphs to explicitly model control dependencies and variable interactions, providing a semantics-aware representation of code.Experiments on the CodeContests+ dataset show that CSSG consistently outperforms existing metrics in distinguishing more similar code from less similar code under both monolingual and cross-lingual settings, demonstrating that dependency-aware graph representations offer a more effective alternative to surface-level or syntax-based similarity measures.

Problem

• code similarity
• semantic relationships
• program dependence graphs
• syntax-based metrics
• surface-level overlap

Innovation

• semantic graph
• program dependence graph
• code similarity
• control dependency
• cross-lingual code analysis