RepoShapley: Shapley-Enhanced Context Filtering for Repository-Level Code Completion
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910139973369856 |
|---|---|
| author | Huo, Yu Zeng, Kun Zhang, Siyu Lu, Yuquan Yang, Cheng Guo, Yifu Tang, Xiaoying |
| author_facet | Huo, Yu Zeng, Kun Zhang, Siyu Lu, Yuquan Yang, Cheng Guo, Yifu Tang, Xiaoying |
| contents | Repository-level code completion benefits from retrieval-augmented generation (RAG). However, controlling cross-file evidence is difficult because chunk utility is often interaction-dependent: some snippets help only when paired with complementary context, while others harm decoding when they conflict. We propose RepoShapley, a coalition-aware context filtering framework supervised by Shapley-style marginal contributions. Our offline labeling module, ChunkShapley, estimates signed per-chunk effects via teacher-forced probing, feeds them into a lightweight surrogate game that captures saturation and interference, computes exact Shapley values for small retrieval sets, and selects a decoding-optimal coalition through bounded post-verification with the frozen generator. The verified keep/drop decisions and retrieval triggers are then distilled into a single model via discrete control tokens. Experiments across benchmarks and backbones show that RepoShapley improves completion quality while reducing harmful context and unnecessary retrieval. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_03378 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RepoShapley: Shapley-Enhanced Context Filtering for Repository-Level Code Completion Huo, Yu Zeng, Kun Zhang, Siyu Lu, Yuquan Yang, Cheng Guo, Yifu Tang, Xiaoying Software Engineering Repository-level code completion benefits from retrieval-augmented generation (RAG). However, controlling cross-file evidence is difficult because chunk utility is often interaction-dependent: some snippets help only when paired with complementary context, while others harm decoding when they conflict. We propose RepoShapley, a coalition-aware context filtering framework supervised by Shapley-style marginal contributions. Our offline labeling module, ChunkShapley, estimates signed per-chunk effects via teacher-forced probing, feeds them into a lightweight surrogate game that captures saturation and interference, computes exact Shapley values for small retrieval sets, and selects a decoding-optimal coalition through bounded post-verification with the frozen generator. The verified keep/drop decisions and retrieval triggers are then distilled into a single model via discrete control tokens. Experiments across benchmarks and backbones show that RepoShapley improves completion quality while reducing harmful context and unnecessary retrieval. |
| title | RepoShapley: Shapley-Enhanced Context Filtering for Repository-Level Code Completion |
| topic | Software Engineering |
| url | https://arxiv.org/abs/2601.03378 |