Chopping Trees: Semantic Similarity Based Dynamic Pruning for Tree-of-Thought Reasoning

2025-10-30

AI Summary

To address excessive computational overhead caused by semantic redundancy in Tree-of-Thought (ToT) reasoning, this paper proposes a dynamic semantic pruning method: during parallel tree search, lightweight semantic similarity computation is employed to cluster and merge equivalent reasoning paths in real time. It introduces, for the first time, an online semantic merging mechanism into the ToT framework, enabling dynamic identification and pruning of redundant branches while preserving reasoning accuracy. Evaluated on GSM8K and MATH500 benchmarks, the method achieves up to 2.3× inference speedup, reduces node exploration by 85–90%, and incurs at most a 5-percentage-point accuracy drop relative to the strongest baseline. The core contribution lies in establishing a real-time semantic deduplication mechanism that jointly optimizes efficiency and fidelity—offering a novel paradigm for efficient complex reasoning with large language models.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Tree-of-Thought (ToT) reasoning boosts the problem-solving abilities of Large Language Models (LLMs) but is computationally expensive due to semantic redundancy, where distinct branches explore equivalent reasoning paths. We introduce Semantic Similarity-Based Dynamic Pruning (SSDP), a lightweight method that, to the best of our knowledge, is the first framework to integrate online semantic merging into parallelized tree search, enabling the clustering and pruning of redundant steps in real time. Across reasoning benchmarks, including GSM8K and MATH500, SSDP achieves up to a 2.3x speedup over state-of-the-art tree-search baselines while maintaining competitive accuracy (typically within 5% of the strongest baseline) and reducing the number of explored nodes by 85-90%, demonstrating a practical approach to efficient, scalable LLM reasoning. The implementation of SSDP is publicly available at https://github.com/kimjoonghokim/SSDP.

Problem

• Reduces computational cost of Tree-of-Thought reasoning by pruning redundant branches.
• Integrates online semantic merging into parallelized tree search for real-time pruning.
• Achieves significant speedup and node reduction while maintaining competitive accuracy.

Innovation

• Semantic similarity-based dynamic pruning for tree-of-thought reasoning
• Online semantic merging integrated into parallelized tree search
• Clustering and pruning redundant steps in real time