Research on the Knowledge Graph Embedding for Rail Fault Text Based on Translation Model

To enhance the automation of urban rail scheduling systems in the face of malfunctions, the current prevalent approach involves constructing intelligent retrieval and question-answering systems based on urban rail fault data knowledge graphs to support dispatchers' decision-making. However, the knowledge graphs within the urban rail sector are currently crafted manually or semi-automatically, potentially neglecting the implicit semantic connections and obscured structural characteristics interlinking various entities. This oversight leads to incomplete rail fault knowledge graphs, adversely affecting the accuracy of intelligent question-answering systems. Consequently, it is essential to train a model to uncover these implicit semantic links and complete the rail fault knowledge graph. This paper proposes a knowledge graph embedding model that integrates relational weight constraints with neighborhood information. The concept of relational hyperplane is incorporated into the model, enabling the embedding vectors of entities and relations to adapt to the complex relational mappings in the urban rail domain. Furthermore, neighborhood information is integrated as supplementary information within the model using a self-attention mechanism, serving to further unearth the implicit semantic connections between rail fault entities, reduce the impact of small sample data on the model's representation capabilities and enhance the model's effectiveness in completing the knowledge graph. Finally, the experiments on a real rail fault dataset are conducted with the proposed model, and the result shows that the model proposed in this paper significantly outperforms other baseline models in triplet classification tasks and link prediction tasks.