A hybrid transformer-GNN framework for social governance and urban service allocation.

The proposed study puts forward an artificial intelligence-based framework to predict the needs of the urban public services and aid resource allocation based on data in the current social governance systems. A hybrid deep learning model is designed by combining a Graph Neural Network (GNN) based on spatial-relational reasoning with a Transformer network to model time-dependent connections, textual complaint semantics and structural relations between service requests, agencies, and locations, and via the joint learning of time-dependent patterns, textual complaint semantics, and structural relationships among service requests, agencies, and locations. In order to deal with the high-dimensional and non-convex problem of hyperparameter tuning on hybrid architectures, an Improved Heap-Based Optimizer (IHBO) is used, using opposition-based learning and chaotic search strategies to improve convergence and global search. The suggested model is tested using the Official Website of the City of New York (NYC 311) Service Requests large-scale data of nearly 12 million records that have mixed temporal, geographic, and categorical variables. It is experimentally proven that the IHBO-optimized Transformer-GNN has an overwhelming performance in comparison to the state-of-the-art baselines with a classification accuracy of 0.938 and lower prediction error by resolution time with Root Mean Square Error equal to 2.18 days, and it is also robust to novel temporal variations and noisy labels. In addition to predictive performance, the suggested model can deliver policy implications to urban governance by making allocation of public service resources more adaptive, equitable, and efficient, which do attest to the utility of hybrid artificial intelligent models in citizen-focused government of any kind.