LoSemB: Logic-Guided Semantic Bridging for Inductive Tool Retrieval

TL;DR

LoSemB addresses the inductive tool retrieval problem in evolving tool repositories by introducing a logic-guided framework that transfers latent logical information to unseen tools. It combines a logic-based embedding alignment (via graph convolution and feature transfer) with a relational augmented retrieval that performs logical pruning and graph-enhanced similarity matching. Empirical results show LoSemB outperforms both zero-shot and training-based baselines in inductive settings while maintaining strong transductive performance, with notable robustness to unseen tools and label noise and improved downstream generation quality. This approach offers a practical path toward robust, scalable tool retrieval for real-world, ever-expanding tool ecosystems.

Abstract

Tool learning has emerged as a promising paradigm for large language models (LLMs) to solve many real-world tasks. Nonetheless, with the tool repository rapidly expanding, it is impractical to contain all tools within the limited input length of LLMs. To alleviate these issues, researchers have explored incorporating a tool retrieval module to select the most relevant tools or represent tools as unique tokens within LLM parameters. However, most state-of-the-art methods are under transductive settings, assuming all tools have been observed during training. Such a setting deviates from reality as the real-world tool repository is evolving and incorporates new tools frequently. When dealing with these unseen tools, which refer to tools not encountered during the training phase, these methods are limited by two key issues, including the large distribution shift and the vulnerability of similarity-based retrieval. To this end, inspired by human cognitive processes of mastering unseen tools through discovering and applying the logical information from prior experience, we introduce a novel Logic-Guided Semantic Bridging framework for inductive tool retrieval, namely, LoSemB, which aims to mine and transfer latent logical information for inductive tool retrieval without costly retraining. Specifically, LoSemB contains a logic-based embedding alignment module to mitigate distribution shifts and implements a relational augmented retrieval mechanism to reduce the vulnerability of similarity-based retrieval. Extensive experiments demonstrate that LoSemB achieves advanced performance in inductive settings while maintaining desirable effectiveness in the transductive setting.

Figures (7)

• Figure 1: Token-based vs. Retrieval-based paradigm.
• Figure 2: (a) Impact of unseen tool ratio on retrieval performance, (b) Comparison of KL divergence distributions between seen and unseen tools, (c) Distribution of tool co-occurrence counts across datasets, (d) Tool overlap percentages between each instruction's tool set and its Top-5 semantically similar instructions. All results are based on ToolBench (I2) and (I3).
• Figure 3: The overall framework of LoSemB.
• Figure 4: Retrieval results (%) of different baselines across varying percentages of unseen tools. Each cell shows the absolute performance and the relative drop ("$\downarrow$") compared to the transductive setting.
• Figure 5: Retrieval results (%) of different baselines under transductive and inductive settings. Clean data indicates no label noise.