Large Language Models Driven Bidding Optimization for Semantics-Aware UAV Assisted Edge Networks

Abstract

Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) provides flexible, real-time computing services to Internet of Things (IoT) devices at network edges. However, existing resource allocation schemes primarily rely on the physical parameters of tasks, while neglecting their semantic priorities. This may cause high-priority tasks to experience a significant drop in performance during network congestion. To address this issue, this paper proposes a large language model (LLM)-driven bidding optimization (LBO) algorithm to enable semantic-aware resource allocation and task offloading. Specifically, we use the natural language understanding capabilities of LLMs to extract semantic features from tasks in order to derive the priority level, total budget, and maximum tolerable latency. Based on these three parameters, we design a hierarchical architecture that utilizes an improved Kelly mechanism for computing power bidding. Then, the system allocates bandwidth and partitions tasks according to the water-filling algorithm and rate-matching principle. Simulation results demonstrate that the proposed LBO algorithm outperforms the traditional Kelly mechanism and the uniform-priority comparison method in terms of social welfare. It successfully enhances the performance of high-priority tasks during periods of severe network congestion.