Jingjing Li

This study proposes a prediction model based on a dynamic spatiotemporal causal graph neural network to address the challenges of complex dynamic dependencies, strong structural correlations, and ambiguous causal relationships in corporate revenue forecasting. The model constructs a time-varying enterprise association graph, where enterprises are represented as nodes and industry or supply chain relationships as edges. A graph convolutional network is used to extract structural dependency features, while a gated recurrent unit captures temporal evolution patterns, achieving joint modeling of structural and temporal features. On this basis, a causal reasoning mechanism is introduced to model and adjust potential influence paths among enterprises. A learnable causal weight matrix is used to describe the strength of economic transmission, suppress spurious correlations, and strengthen key causal paths. The model also employs multi-scale temporal aggregation and attention fusion mechanisms to dynamically integrate multidimensional information, enhancing adaptability to both long-term trends and short-term fluctuations. Experimental results show that the proposed model outperforms mainstream methods in multiple metrics, including MSE, MAE, MAPE, and RMAE, verifying its effectiveness in capturing corporate revenue dynamics, modeling economic causal dependencies, and improving prediction accuracy. This study establishes a unified framework that integrates spatiotemporal dependency modeling with causal structure reasoning, providing new insights and methodological foundations for intelligent forecasting in complex economic systems.

Fraudulent financial reporting continues to undermine market integrity and threaten stakeholder interests globally. Conventional analytical methodologies demonstrate limited effectiveness in recognizing intricate temporal relationships and nuanced irregularities characteristic of accounting manipulation. This research introduces an enhanced recurrent neural architecture incorporating multi-head attention layers with bidirectional LSTM components, specifically engineered for identifying accounting anomalies. Our architectural design combines attention-weighted feature selection with dual-directional temporal processing to simultaneously extract prolonged sequential patterns and pivotal indicators spanning multiple reporting cycles. Through rigorous evaluation using a dataset of 2,500 corporate entities across 12 reporting cycles, our methodology achieves 93.87% classification accuracy, 92.41% positive predictive value, 91.56% sensitivity, and 91.98% harmonic mean of precision-recall. These metrics surpass conventional statistical methods and standard deep architectures by substantial margins. The attention weight distributions offer transparent interpretability, revealing which specific accounting ratios and temporal windows contribute decisively to fraud predictions, thereby enhancing practical deployment feasibility in professional audit environments.