Graph Neural Network-Based Organizational Decision Optimization Algorithm for Dynamic Resource Allocation in Enterprises

Abstract

Resource allocation remains a critical consideration within the domain of business administration due to changing demands, interdependence between organizational units, and conflicts among objectives. Traditional optimization techniques, including linear programming, heuristic algorithms, and rule-based systems, have limited ability to model the relationship dynamics inherent in modern organizations. In this paper, we propose GNN-ODOA (Graph Neural Network-Based Organizational Decision Optimization Algorithm), a new technique that represents organizations using heterogeneous attributed graphs and uses message-passing neural networks for transferring contextual data through organizational structures. The proposed model employs a temporal attention module to learn time-dependent demand trends from allocation histories, a multi-objective reinforcement learning algorithm to optimize trade-offs between cost-effectiveness, processing capacity, and fairness requirements, and a conflict-resolution mechanism to solve conflicts between simultaneous resource allocations. Evaluations of the proposed algorithm, GNN-ODOA, will be conducted using three real-world enterprise environments, including one multinational manufacturer, one large-scale IT service provider, and one distributed logistics company. Our experiments show that our model achieves 23.7% higher resource utilization compared to the best baseline, 41.2% lower allocation latency, and Pareto-optimal allocation in terms of constraints. The ablation study reveals the role of all components of the architecture, and scalability test verifies the linear increase in the time complexity to process inference when increasing the number of nodes up to 10,000.