A Penetrative Multidimensional Data Analytics Model for Complex Relationship Mining over Knowledge Graphs

This study proposes a deep multidimensional data analytics framework for extracting intricate relationships from knowledge graphs, which tackles the challenge of discovering hidden connections in heterogeneous and high-dimensional datasets. The proposed method unifies three principal elements: Dynamic Meta-Path Penetration, Nested Subgraph Extraction, and Tensor-Graph Fusion, which together permit a structured investigation of hidden connections. Dynamic Meta-Path Penetration applies reinforcement learning to traverse the graph, directed by a reward system prioritizing informative routes. Nested Subgraph Extraction hierarchically aggregates multi-hop dependencies by employing Graph Neural Networks, which identifies structural patterns within localized subgraphs. Tensor-Graph Fusion performs joint factorization on the knowledge graph adjacency tensor and multidimensional data tensors, thereby merging structural and attribute-based information within a common latent space. The PPA-GNN layer coordinates these elements by traversing the graph, eliminating unnecessary connections, and merging cross-modal attributes, thus producing embeddings that capture intricate relationships. Additionally, the penetration depth is established as a metric to measure the minimal distance needed to uncover hidden relationships. Experiments on benchmark datasets show our model achieves better performance than state-of-the-art methods in relationship mining tasks, especially in cases with sparse or noisy data. The framework’s ability to integrate heterogeneous data sources and dynamically adapt to graph structures makes it suitable for applications in recommendation systems, biomedical discovery, and social network analysis. This study propels the discipline forward by introducing a cohesive framework for penetrative analytics, which connects graph-based and tensor-based approaches.