A Human-in-Multi-Agent-Loop Intent Refinement Method for Task Offloading in Mobile Edge Computing

In mobile edge computing (MEC), intent recognition plays a pivotal role in task offloading. To address this challenge, this paper proposes a Human-in-Multi-Agent-Loop (HIMAL) intent refinement method for MEC that synergistically integrates multi-agent collaboration with human judgment to enhance resource allocation and task scheduling efficiency, thereby facilitating the advancement of the Industrial Internet of Things (IIoT). The proposed HIMAL system enhances AI decision-making through continuous integration of human feedback into the AI’s learning process, thereby enhancing both the precision of AI outputs and contextual awareness. Experimental results demonstrate that our approach achieves significant performance improvements by embedding human guidance within large language model (LLM) workflows, where human operators serve as contextual interpreters to refine AI responses.