Multi-modal Knowledge Unified Representation Learning Method Adapted to Dynamic Data Update for Power System
Ting Guo, Hui Xiang, Di Liu, Dawei Lu, Zhen Qiu · 2024
In typical power industry applications, the multimodal knowledge graph faces the challenges of difficulty in integrating and utilizing modal information, and difficulty in dynamically adjusting knowledge representation as the graph is updated. To address this issue, this paper proposes a unified representation learning method for multi-modal knowledge graphs suitable for typical power business applications. Firstly, a unified representation learning framework is proposed for multi-modal knowledge graphs tailored for typical power businesses. Secondly, a knowledge representation learning technique that integrates entity descriptions and hierarchical information is introduced. Thirdly, a knowledge representation learning technique based on attention mechanism is proposed to integrate solid image information. Finally, a unified representation updating method is presented for multimodal knowledge graphs with dynamic data updates. The experimental results illustrate that the proposed method can efficiently fusion and update the power multi-modal knowledge, and provide unified support for knowledge-driven tasks such as information retrieval, interaction and reasoning based on knowledge graph.