Integrating Causal Inference and Graph Attention for Structure-Aware Data Mining

Abstract

This paper addresses the limitations of traditional data mining methods in causal modeling and structural awareness. It proposes a causal-enhanced data mining algorithm that integrates causal inference with a graph attention mechanism. The method is grounded in causal structure learning. It first constructs a causal graph among variables based on conditional independence constraints. This causal graph is then embedded into a graph neural network framework to achieve structured representation of causal relationships and high-order information aggregation. During the graph modeling process, a causal weight control mechanism is introduced to regulate the strength of information flow between nodes. This allows the model to adaptively capture key causal paths and significant dependencies. At the same time, an attention mechanism assigns weights to neighboring nodes, enhancing the model's ability to identify important factors within complex structures. In the optimization phase, the model jointly uses task-specific loss and a causal consistency regularization term to improve its ability to fit the true causal structure. To verify the effectiveness of the proposed method, experiments are conducted on datasets with clearly defined causal structures. The evaluation focuses on multiple aspects, including structural error, inference accuracy, and generalization capability. Comparative analyses are performed against several representative baseline methods. The results demonstrate that the proposed approach achieves superior performance in structural recovery, causal path identification, and predictive accuracy. These findings highlight the powerful capacity of combining causal modeling with graph structure learning for modeling complex systems.