Code Clone Detection via an AlphaFold-Inspired Framework

2025-07-20

AI Summary

Existing code clone detection methods struggle to simultaneously achieve strong semantic modeling capability and cross-language generalizability. To address this, we propose the first language-agnostic semantic clone detection framework inspired by AlphaFold’s protein structure prediction paradigm—specifically, its integration of multiple sequence alignment (MSA) and attention mechanisms. Our approach constructs tokenized sequences, performs cross-language MSA alignment, employs a modified attention-based encoder to capture deep semantic representations, and incorporates interactive similarity computation for binary classification. Evaluated on three multilingual benchmark datasets, our method significantly outperforms all baselines, achieving both higher accuracy (+4.2% F1 score) and practical efficiency (3.1× faster inference). This work establishes a scalable, semantics-driven paradigm for cross-language software maintenance and vulnerability discovery.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Computer Vision: Language and Vision

Application Category

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

Abstract

Code clone detection, which aims to identify functionally equivalent code fragments, plays a critical role in software maintenance and vulnerability analysis. Substantial methods have been proposed to detect code clones, but they fall short in capturing code semantics or relying on language-specific analyzers. Inspired by the remarkable success of AlphaFold in predicting three-dimensional protein structures from protein sequences, in this paper, we leverage AlphaFold for code clone detection based on the insight that protein sequences and token sequences share a common linear sequential structure. In particular, we propose AlphaCC, which represents code fragments as token sequences to ensure multi-language applicability and adapts AlphaFold's sequence-to-structure modeling capability to infer code semantics. The pipeline of AlphaCC goes through three steps. First, AlphaCC transforms each input code fragment into a token sequence and, motivated by AlphaFold's use of multiple sequence alignment (MSA) to enhance contextual understanding, constructs an MSA from lexically similar token sequences. Second, AlphaCC adopts a modified attention-based encoder based on AlphaFold to model dependencies within and across token sequences. Finally, unlike AlphaFold's protein structure prediction task, AlphaCC computes similarity scores between token sequences through a late interaction strategy and performs binary classification to determine code clone pairs. Comprehensive evaluations on three language-diverse datasets demonstrate AlphaCC's applicability across multiple programming languages. On two semantic clone detection datasets, it consistently outperforms all baselines, showing strong semantic understanding. Moreover, AlphaCC maintains competitive efficiency, enabling practical usage in large-scale clone detection tasks.

Problem

• Detects code clones across multiple programming languages
• Improves semantic understanding of code fragments
• Enhances efficiency for large-scale clone detection

Innovation

• Adapts AlphaFold for multi-language code clone detection
• Uses token sequences and MSA for semantic understanding
• Late interaction strategy for similarity scoring