Dynamic Graph Learning for Core Data Asset Discovery Using Data Lineage Graphs

The rapid growth of the digital economy has made the accurate identification of Core Data Assets (CDAs) increasingly important for enterprise data governance. Existing approaches, however, often rely on static, homogeneous dependency modeling or node-centric analysis, limiting their ability to capture complex value propagation patterns and the structural importance of data in enterprise business processes. To address these limitations, this study proposes a graph learning framework with dynamic (structure-adaptive) edge weighting for automated CDA discovery using data lineage graphs. The framework first introduces a dynamic edge-weighting mechanism that integrates node influence and relationship scarcity to represent heterogeneous value propagation between data entities. The framework then constructs a comprehensive feature representation by combining global influence, local topological characteristics, and functional connectivity patterns, enabling the model to capture different aspects of data-table importance. Based on these representations, we develop an ensemble learning model with group-based cross-validation and automated hyperparameter optimization for robust CDA classification. Experiments on five real-world lineage graphs from a single enterprise data warehouse demonstrate the framework’s effectiveness. The model achieves average AUC and F1 scores of 0.926 and 0.753, respectively, and outperforms the evaluated baseline methods in terms of F1-score and Precision; the strongest competing baseline attains higher AUC and Recall but substantially lower Precision and F1-score. The results show that the multidimensional structural features provide the main share of the improvement, that the dynamic weighting mechanism contributes a modest additional benefit, and that the approach is effective for lineage graphs of moderate scale in enterprise data governance.