Project-wise Comparison of Software Birthmarks Using Weighted Partial Similarity

2013-01-10 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Software Development & Engineering, Cybersecurity & Data Privacy, Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

A project-wise software birthmark comparison framework addresses challenges in detecting code plagiarism, particularly partial reuse and incidental similarity in large-scale software. This framework, based on symmetric aggregation of module-level similarities, introduces a weighting scheme that prioritizes larger modules and a partial similarity method focusing on the top fraction of highly similar module pairs. Evaluated on 35 open-source Java projects across ten categories, treating different versions as reuse cases, the method consistently outperformed existing approaches. It achieved robust and stable detection of partial code reuse at the project level, with optimal performance observed using small comparison scopes (1-5%), k-gram sizes of 3-4, and edit distance as the module-wise similarity function. The dataset and experimental artifacts are publicly available.

Key takeaway

For software engineers or research scientists tasked with detecting code reuse in large Java projects, you should adopt a project-wise birthmark comparison framework. Implement size-based weighting for modules and focus on the top 1-5% of highly similar module pairs, ideally using edit distance for module similarity. This approach significantly improves detection accuracy and stability, reducing false positives from small, generic modules and effectively identifying partial code reuse. Consider k-gram sizes of 3 or 4 for optimal performance.

Key insights

Project-wise software birthmark comparison requires weighting module similarities and focusing on top matches to detect partial reuse and mitigate incidental similarity.

Principles

• Symmetric aggregation provides a strong baseline for project similarity.
• Weighting modules by size reduces false positives from incidental similarity.
• Partial similarity enhances reuse detection by focusing on relevant module pairs.

Method

The framework uses symmetric aggregation, applies logarithmic weighting to module birthmark sizes, and employs partial similarity by selecting the top α% of weighted module-pair similarities.

In practice

• Filter external and small modules (LLC ≤ 30) for cleaner data.
• Use k-gram sizes k ∈ {3,4} for balanced performance.
• Prioritize edit distance for module-wise similarity calculations.

Topics

• Software Birthmarks
• Code Plagiarism Detection
• Project-wise Similarity
• Weighted Similarity
• Partial Code Reuse
• Java Software Analysis
• Edit Distance

Code references

• tamada/pochi
• ou-salab-fedorov/project-wise_comparison_of_software_birthmarks

Best for: AI Scientist, Research Scientist, Software Engineer

Related on AIssential

• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — RARE framework improves RAG evaluation by accounting for document redundancy in high-similarity real-world corpora.
• Faithful by Construction: Claim-Anchored Attribution for Multi-Document Summarization — A modular, claim-anchored summarization framework inherently builds fine-grained attribution and faithfulness into the generation process.
• IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources — Semantic deduplication and domain-balanced pretraining significantly enhance Persian PLM performance across diverse NLU tasks.
• Proposal and study of statistical features for string similarity computation and classification — Visual computing features, COM and RLM, offer superior, language-agnostic string similarity computation.
• Anchor Detection for RAG: Parallel Detectors, Then One LLM Call at the End — Enterprise RAG retrieval combines parallel keyword and embedding detectors, aggregating results for a single, auditable LLM ranking.
• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — Existing RAG benchmarks fail in redundant, real-world corpora, necessitating a new evaluation framework.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.