Detection and Recognition of Accounting Information Disclosure Behavior in Educational Institutions Based on Graph Neural Networks

To enhance the detection of financial misconduct in educational institutions, this paper proposes Edu-DiscNet, a graph neural network model that integrates structured indicators with textual semantics. The model incorporates RGCN++ to represent multi-relational structures, combines T-GAT and GRU to capture temporal evolution features, and adopts a multi-task framework combining node anomaly detection and graph-level disclosure rating for comprehensive evaluation. Experimental results demonstrate that Edu-DiscNet significantly outperforms existing models in terms of precision, F1-score, and AUC, and exhibit strong consistency in disclosure-level ranking. Ablation studies further confirm the critical roles of each module in improving robustness and discriminative performance.