Deep Learning Empowered Enterprise Knowledge Graph with Attention Mechanism

Enterprise knowledge graphs (EKGs) are pivotal in structuring and analyzing vast amounts of enterprise data, yet conventional construction methods struggle to efficiently capture complex relationships and dynamic enterprise contexts. This paper proposes a Deep Learning (DL)-based enterprise knowledge graph framework that integrates transformer-based architectures, graph attention networks (GATs), and reinforcement learning to enhance the construction, refinement, and querying of EKGs. Specifically, we employ a business-enhanced RoBERTa (BERTO) model for entity and relation extraction from unstructured data, a graph attention network for refining edge weights, and a reinforcement learning agent to adaptively update relationships based on user feedback. Additionally, a query-aware attention mechanism is incorporated for context-sensitive knowledge retrieval. Simulation results demonstrate that the proposed scheme outperforms conventional knowledge graph (GK) and deep learning (DL) models in predictive accuracy, especially under varying signal-to-noise ratio (SNR) conditions. Numerical comparisons reveal that at 10 dB SNR, the proposed scheme achieves a prediction accuracy of 0.74, surpassing the conventional GK (0.49) and conventional DL (0.34) methods. These results underscore the effectiveness of the proposed framework in improving accuracy, adaptability, and scalability in enterprise knowledge management.