Research map / Label-efficient and robust learning / Recognition and long-tail
CHICOT: A Developer-Assistance Toolkit for Code Search with High-Level Contextual Information
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents CHICOT, a VSCode-integrated code search toolkit that retrieves function-plus-README-context pairs from GitHub to match high-level project context in developer queries.
Problem
Existing semantic code search matches only code-level specifications (functionality, logic, implementation) and ignores high-level project context such as purpose or domain, which developers need when reusing code.
Approach
Crawl GitHub functions paired with a context snippet extracted from each repository's README (LEAD-k summary), then rank function-context pairs by a weighted combination of query-function and query-context cosine similarities, with the weight w user-configurable.
Key results
- CHICOT achieved precision@10 of 6.4% on 25 manually crafted context-dependent queries.
- Crawler collected 20,000 GitHub repositories (over 100 stars) yielding six million functions.
- Case study (Figure 3) shows retrieved pairs satisfying both code- and context-level requirements for example queries.
Surprising findings
- Only 6.4% of the top-10 retrieved pairs were judged accurate on both code and context alignment, showing the task is hard.
Limitations
- Low precision@10 (6.4%) on context-dependent queries.
- Evaluation limited to 25 manually crafted queries and a qualitative case study.
- Context is limited to the repository README with LEAD-k extraction, shared across functions.
Experimental setup
Datasets: CodeSearchNet MS MARCO GitHub repositories (20,000 with over 100 stars)
Benchmarks: Manually crafted context-dependent queries (25 queries)
Metrics: precision@10
Baselines: CodeSearchNet
Builds on: CodeSearchNet
Closest papers by meaning
- CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval (CVPR 2026 · similarity 0.85)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.85)
- Simple Baselines are Competitive with Code Evolution (NeurIPS 2026 · similarity 0.84)
- GitTaskBench: A Benchmark for Code Agents Solving Real-World Tasks Through Code Repository Leveraging (AAAI 2026 · similarity 0.83)
- Repoformer: Selective Retrieval for Repository-Level Code Completion (ICML 2024 · similarity 0.83)
- Towards Practical Defect-Focused Automated Code Review (ICML 2025 · similarity 0.83)
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.