Tool Retrieval Bridge: Aligning Vague Instructions with Retriever Preferences via Bridge Model

2026-04-09

AI Summary

This study addresses the challenge posed by ambiguous user instructions in real-world scenarios, which significantly degrades the performance of existing tool retrieval methods that rely on detailed queries. To systematically investigate this issue, the authors introduce VGToolBench, a benchmark designed to simulate realistic ambiguous queries, and propose Tool Retrieval Bridge (TRB), a lightweight and efficient approach that leverages large language models to rewrite vague instructions into more specific forms compatible with standard retrieval systems such as BM25. Experimental results demonstrate that TRB substantially improves retrieval effectiveness on VGToolBench, boosting BM25’s NDCG from 9.73 to 19.59—a relative improvement of 111.51%—and consistently enhances performance across all baseline retrievers.

Abstract

Tool learning has emerged as a promising paradigm for large language models (LLMs) to address real-world challenges. Due to the extensive and irregularly updated number of tools, tool retrieval for selecting the desired tool subset is essential. However, current tool retrieval methods are usually based on academic benchmarks containing overly detailed instructions (e.g., specific API names and parameters), while real-world instructions are more vague. Such a discrepancy would hinder the tool retrieval in real-world applications. In this paper, we first construct a new benchmark, VGToolBench, to simulate human vague instructions. Based on this, we conduct a series of preliminary analyses and find that vague instructions indeed damage the performance of tool retrieval. To this end, we propose a simple-yet-effective Tool Retrieval Bridge (TRB) approach to boost the performance of tool retrieval for vague instructions. The principle of TRB is to introduce a bridge model to rewrite the vague instructions into more specific ones and alleviate the gap between vague instructions and retriever preferences.We conduct extensive experiments under multiple commonly used retrieval settings, and the results show that TRB effectively mitigates the ambiguity of vague instructions while delivering consistent and substantial improvements across all baseline retrievers. For example, with the help of TRB, BM25 achieves a relative improvement of up to 111.51%, i.e., increasing the average NDCG score from 9.73 to 19.59. The source code and models are publicly available at https://github.com/kfchenhn/TRB.

Problem

• tool retrieval
• vague instructions
• large language models
• retriever preferences
• instruction ambiguity

Innovation

• Tool Retrieval
• Vague Instructions
• Bridge Model
• Instruction Rewriting
• VGToolBench