GPT-Augmented Bayesian Reinforcement Learning Framework for Multiobjective Supplier Selection

In today's volatile geopolitical environment and heightened emphasis on sustainability, effective supplier selection must simultaneously handle cost, delivery risks, and ESG considerations. This paper proposes a GPT-Augmented Bayesian Reinforcement Learning (i-SUP) Framework, which integrates 1) GPT to extract real-time risk signals from unstructured text (news, social media), 2) Bayesian-BWM to capture expert uncertainty and produce robust multi-criteria weights, 3) Bayesian Belief Networks (BBNs) for continuously updated disruption probabilities, 4) Reinforcement Learning (RL) for dynamic monthly or weekly order allocation, and 5) NSGA-II for long-horizon multi-objective contract planning. By combining semantic risk detection with Bayesian updates and RL-based adaptive decision-making, i-SUP dynamically adjusts to emergent risks (e.g., tariffs, labor unrest), while concurrently balancing ESG imperatives and cost efficiency. Empirical validation in the semiconductor industry—characterized by tight geopolitical sensitivity and high ESG demands—shows that i-SUP significantly reduces disruptions and ESG incidents relative to static or cost-only methods. Moreover, ablation analyses confirm that removing any single module (GPT, BBN, RL, or NSGA-II) undermines performance, demonstrating the necessity of a fully integrated pipeline. The findings underscore i-SUP's ability to enhance supplier resilience and sustainability in a wide range of globalized supply networks that face evolving textual risk signals and multi-dimensional objectives. ESG trade-offs, enabling boards and procurement leaders to align operational policies with strategic sustainability goals. Overall, this paper presents a policy-relevant approach that informs how top management can leverage artificial intelligence in day-to-day supplier decisions and long-term contract negotiations, thereby advancing both financial outcomes and corporate social responsibility in high-stakes, rapidly changing supply networks.