Event-CausNet: Unlocking Causal Knowledge from Text with Large Language Models for Reliable Spatio-Temporal Forecasting

2025-11-16

AI Summary

Traditional spatiotemporal graph neural networks (GNNs) rely heavily on historical correlation modeling, leading to a sharp decline in prediction reliability under unexpected events (e.g., traffic accidents). To address this, we propose the first traffic forecasting framework integrating causal inference with spatiotemporal modeling. Our method leverages large language models to extract causal knowledge from unstructured event texts and constructs a causal knowledge base grounded in Average Intervention Effect (AIE). We design a causal attention mechanism to inject this knowledge into the model and adopt a dual-stream architecture that jointly fuses spatiotemporal graph convolutions, LSTMs, and causal representations. This approach significantly enhances robustness under disruptions, improves interpretability, and strengthens cross-scenario generalizability. Evaluated on real-world traffic datasets, our method achieves up to a 35.87% reduction in MAE, outperforming existing state-of-the-art approaches.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

While spatio-temporal Graph Neural Networks (GNNs) excel at modeling recurring traffic patterns, their reliability plummets during non-recurring events like accidents. This failure occurs because GNNs are fundamentally correlational models, learning historical patterns that are invalidated by the new causal factors introduced during disruptions. To address this, we propose Event-CausNet, a framework that uses a Large Language Model to quantify unstructured event reports, builds a causal knowledge base by estimating average treatment effects, and injects this knowledge into a dual-stream GNN-LSTM network using a novel causal attention mechanism to adjust and enhance the forecast. Experiments on a real-world dataset demonstrate that Event-CausNet achieves robust performance, reducing prediction error (MAE) by up to 35.87%, significantly outperforming state-of-the-art baselines. Our framework bridges the gap between correlational models and causal reasoning, providing a solution that is more accurate and transferable, while also offering crucial interpretability, providing a more reliable foundation for real-world traffic management during critical disruptions.

Problem

• Extracting causal knowledge from unstructured text for reliable forecasting
• Addressing spatio-temporal model failures during non-recurring disruptive events
• Bridging the gap between correlational models and causal reasoning

Innovation

• Uses LLM to quantify unstructured event reports
• Builds causal knowledge base via treatment effects
• Injects causal knowledge via novel attention mechanism