Research map / Vision-language and multimodal / Human and embodied modeling
NS3: Neuro-symbolic Semantic Code Search
Shushan Arakelyan, Anna Hakhverdyan, Miltiadis Allamanis, Luis Antonio Garcia, Christophe Hauser, Xiang Ren
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use NS³'s parse-driven modular re-ranking instead of GraphCodeBERT* re-ranking | CodeSearchNet · CodeSearchNet test | MRR: 0.883 -> 0.924 | Improved |
| Use NS³ instead of GraphCodeBERT | CodeSearchNet with 10K training examples · CodeSearchNet-10K test | MRR: 0.786 -> 0.826 | Improved |
| Use NS³ instead of GraphCodeBERT | CodeSearchNet with 5K training examples · CodeSearchNet-5K test | MRR: 0.773 -> 0.823 | Improved |
| Use NS³ instead of GraphCodeBERT* re-ranking | CoSQA-trained model · WebQueryTest | MRR: 0.462 -> 0.551 | Improved |
| Use NS³ with CodeBERT fallback for unparsable examples instead of GraphCodeBERT | CoSQA-trained model · CoSQA full test set | MRR: 0.367 -> 0.412 | Improved |
| Add action-module pretraining before end-to-end training (ablation) | NS³ · CodeSearchNet test | MRR | Improved |
| Add both module pretraining procedures before end-to-end training (ablation) | NS³ · CodeSearchNet test | MRR | Improved |
| Add end-to-end training after module pretraining (ablation) | NS³ · CodeSearchNet test | MRR | Improved |
| Normalize module output scores (ablation) | NS³ · CodeSearchNet and CoSQA | MRR | Improved |
| Normalize only action-module or only entity-discovery output rather than both (ablation) | NS³ · CodeSearchNet and CoSQA | MRR | No clear effect |
| Replace dot-product similarity with L2 distance or weighted cosine similarity (ablation) | NS³ · CodeSearchNet test | MRR | No clear effect |
Closest papers by meaning
- CodePlan: Unlocking Reasoning Potential in Large Language Models by Scaling Code-form Planning (ICLR 2025 · similarity 0.85)
- Towards A Generalist Code Embedding Model Based On Massive Data Synthesis (NeurIPS 2025 · similarity 0.85)
- CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval (CVPR 2026 · similarity 0.84)
- CodeIO: Condensing Reasoning Patterns via Code Input-Output Prediction (ICML 2025 · similarity 0.84)
- CPRet: A Dataset, Benchmark, and Model for Retrieval in Competitive Programming (NeurIPS 2025 · similarity 0.84)
- CodeSteer: Symbolic-Augmented Language Models via Code/Text Guidance (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
