SkillGATE: Gate-Aware Monte Carlo Tree Search for Skill Retrieval

2026-10-04

AI Summary

This study addresses the accuracy degradation in large-scale skill library retrieval caused by semantic interference, local optima traps, and early routing errors. To overcome these challenges, this work proposes SkillGATE, a framework that formulates retrieval as an adaptive information foraging process. It introduces a novel gate-aware Monte Carlo Tree Search (MCTS) mechanism, where G-PUCT guides action selection to coordinate region-level navigation with skill-level filtering. Furthermore, the framework incorporates a graph-preserving hierarchical index construction technique to optimize search statistics for guiding retrieval decisions. Extensive experiments across six benchmarks demonstrate that the proposed method significantly enhances the performance of various backbone models, achieving a 16.3% improvement in R@1 over the strongest baseline.

Abstract

Skill Retrieval (SR) aims to identify the most relevant skills from external skill libraries, and becomes increasingly challenging as libraries grow in scale and diversity. Existing methods either rank skills independently or rely on predefined graph propagation and hierarchical routing, making them vulnerable to semantic distractors, local trapping, and early routing errors. We formulate SR as an adaptive information-foraging process that coordinates region-level navigation with skill-level selection according to the utility and uncertainty observed during search. Based on this formulation, we propose SkillGATE, a graph-guided hierarchical retrieval framework with Gate-Aware Monte Carlo Tree Search (MCTS). SkillGATE constructs a graph-preserving hierarchical index and performs adaptive retrieval through selection, expansion, simulation, and backpropagation. G-PUCT guides action selection, expansion explores new regions, simulation evaluates candidate skills, and backpropagation updates search statistics. Experiments on six SR benchmarks show that SkillGATE consistently improves diverse retrieval and reranking backbones, achieving a 16.3\% improvement in overall R@1 over the strongest retriever-based baseline. Our code is available at https://github.com/Edwinbe/SkillGATE-v1/.

Problem

• Skill Retrieval
• Semantic Distractors
• Local Trapping
• Early Routing Errors

Innovation

• Skill Retrieval
• Monte Carlo Tree Search
• Hierarchical Index
• Information Foraging
• G-PUCT