--- title: "Graph-Based Chain-of-Thought Pruning for Reducing Redundant" description: "An external LLM then incrementally builds a DAG whose nodes are labelled progress or review and… 6 references, cited by 1 paper here." authors: ["Hongyuan Yuan", "Xinran He", "Run Shao", "Bolei He", "Xianwei Xue", "Mengke Chen", "Qiutong Pan", "Haiwei Wang", "Haifeng Li"] year: 2026 venue: "Findings of the Association for Computational Linguistics: ACL 2026" source: "https://aclanthology.org/2026.findings-acl.281/" doi: "10.18653/v1/2026.findings-acl.281" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/yuan-2026-graph-based-chain-thought-pruning/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yuan-2026-graph-based-chain-thought-pruning/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/13/) / Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs # Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs **Authors:** Hongyuan Yuan, [Xinran He](https://md.research.elamur.ai/free/en/authors/xinran-he/), [Run Shao](https://md.research.elamur.ai/free/en/authors/run-shao/), [Bolei He](https://md.research.elamur.ai/free/en/authors/bolei-he/), [Xianwei Xue](https://md.research.elamur.ai/free/en/authors/xianwei-xue/), [Mengke Chen](https://md.research.elamur.ai/free/en/authors/mengke-chen/), [Qiutong Pan](https://md.research.elamur.ai/free/en/authors/qiutong-pan/), [Haiwei Wang](https://md.research.elamur.ai/free/en/authors/haiwei-wang/) and Haifeng Li · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.281/) · [PDF](https://aclanthology.org/2026.findings-acl.281.pdf) · [DOI 10.18653/v1/2026.findings-acl.281](https://doi.org/10.18653/v1/2026.findings-acl.281) · [OpenAlex](https://openalex.org/W7166804535) **Citations:** 0 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/yuan-2026-graph-based-chain-thought-pruning/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary Reasoning LLMs trained with reinforcement learning often overthink and produce long chains of thought. The authors attribute much of this waste to unproductive self-reflection of two kinds. The model either checks trivial steps across the board (Indiscriminate Reflection) or keeps re-confirming results it has already verified (Repetitive Reflection). To target these behaviours, they split each CoT into chunks at cue words such as "wait". An external LLM then incrementally builds a DAG whose nodes are labelled progress or review and linked by logical dependencies. Two rules prune the graph: - **Branch-level pruning** drops review nodes that have few descendants. - **Depth-level pruning** drops review nodes that appear late in the reasoning. The pruned traces feed a three-step training pipeline: cold-start SFT, then DPO that favours concise correct trajectories over verbose ones, then GRPO with a length penalty. Across benchmarks, average reasoning tokens fall by about 42%, with accuracy matched or improved against efficiency baselines. This suggests that structure-aware removal of reflections is a practical way to cut inference cost. ## Key points - The graph labels each node as progress (moves the solution forward) or review (checks, restates or backtracks). Only review nodes are candidates for removal, using a descendant-count threshold k and a depth cutoff ratio m. - Compared baselines include Light-R1, AdaptThink, TokenSkip, Skywork-OR1 and O1-Pruner. The headline result is roughly 42% fewer reasoning tokens with accuracy maintained or improved. - The pruning ablation used SFT on DeepSeek-R1-Distill-Qwen-7B: - **Selective pruning:** 42.50% on AIME24 (≈10.4k tokens) and 34.17% on AIME25. - **No pruning:** 42.08% (≈12.7k tokens) and 33.75%. - Selective pruning also beat random pruning and removing every review node, so some reflections are worth keeping. - A threshold sweep (k = 1–3, m = 0.7–0.9) showed stable behaviour. Raising k prunes more aggressively and can lower accuracy, while varying m gives a smoother accuracy–length trade-off. - After training, reflection words ("wait", "hmm", "check") occur less often and "therefore" more often. The long tail of very lengthy responses on AIME24 also shrinks. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [Baidu](https://md.research.elamur.ai/free/en/companies/baidu/) — Company, Beijing, China · 74 works · 15,263 citations - [Central South University](https://md.research.elamur.ai/free/en/universities/central-south-university/) — University, Changsha, China · 15 works · 582 citations 📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works 🧭 Topics (3): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,320 works · [🧮 Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) · 1,426 works · [🪶 Efficient models & compression](https://md.research.elamur.ai/free/en/topics/efficient-models/) · 242 works ## Cite Cite as: Hongyuan Yuan, Xinran He, Run Shao, Bolei He, Xianwei Xue, Mengke Chen, Qiutong Pan, Haiwei Wang and Haifeng Li. 2026. Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs. Findings of the Association for Computational Linguistics: ACL 2026. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{yuan2026graph, title = {{Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs}}, author = {Hongyuan Yuan, Xinran He, Run Shao, Bolei He, Xianwei Xue, Mengke Chen, Qiutong Pan, Haiwei Wang and Haifeng Li}, booktitle = {Findings of the Association for Computational Linguistics: ACL 2026}, year = {2026}, doi = {10.18653/v1/2026.findings-acl.281}, url = {https://doi.org/10.18653/v1/2026.findings-acl.281} } ``` - [← previous Grammar as Control: Modular Language Generation for the Long Tail](https://md.research.elamur.ai/free/en/papers/nakashole-2026-grammar-as-control-modular-language/) - [next → Gretino: A Greek and Latin Dataset to Benchmark Retrieval Systems in Classical Languages](https://md.research.elamur.ai/free/en/papers/toyin-2026-gretino-greek-latin-dataset-benchmark/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/13/) ## References 6 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Pranjal Aggarwal and Sean Welleck. 2025. [L1: Controlling how long a reasoning model thinks with reinforcement learning](https://md.research.elamur.ai/free/en/papers/aggarwal-2025-l1-controlling-long-reasoning-model/). In Second Conference on Language Modeling. 2. AIME. 2024. American invitational mathematics examination. [https://huggingface.co/datasets/](https://huggingface.co/datasets/) HuggingFaceH4/aime_2024. [↗](https://huggingface.co/datasets/) 3. AIME. 2025. American invitational mathematics examination. [https://huggingface.co/datasets/](https://huggingface.co/datasets/) opencompass/AIME2025. [↗](https://huggingface.co/datasets/) 4. AMC. 2023. American mathematics competitions. [https://huggingface.co/datasets/](https://huggingface.co/datasets/) math-ai/amc23. [↗](https://huggingface.co/datasets/) 5. Daman Arora and Andrea Zanette. 2025. Training language models to reason efficiently. Preprint, arXiv:2502.04463. [find](https://scholar.google.com/scholar?q=Daman+Arora+and+Andrea+Zanette%2E+2025%2E+Training+language+models+to+reason+efficiently%2E+Preprint%2C+arXiv%3A2502%2E04463%2E) 6. Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, and Dong Yu. 2025a. Do NOT think that much for 2+3=? On the overthinking of long reasoning models. In Proceedings of the 42nd International… [find](https://scholar.google.com/scholar?q=Xingyu+Chen%2C+Jiahao+Xu%2C+Tian+Liang%2C+Zhiwei+He%2C+Jianhui+Pang%2C+Dian+Yu%2C+Linfeng+Song%2C+Qiuzhi+Liu%2C+Mengfei+Zhou%2C+Zhuosheng+Zhang%2C+Rui+Wang%2C+Zhaopeng+Tu%2C+Haitao+Mi%2C+and+Dong+Yu%2E+2025a%2E+Do+NOT+think+that+m) *Updated 31 October 2026*