A Unified Framework for Attention-Driven and Correlation-Enhanced Relation Extraction in Underground Space

With the development of smart cities, unmanned systems and artificial intelligence have been widely applied in the management of urban underground spaces. The operation and maintenance of environments such as subways and underground shopping malls require efficient information management and accurate relation extraction techniques. However, due to the heterogeneous, dispersed, and semantically ambiguous nature of underground space data, traditional management methods often prove inadequate. This paper proposes a unified relation extraction framework that integrates attention mechanisms along with a correlation enhancement module. By leveraging unmanned systems for automating the processing of multi-scenario data, the framework effectively addresses data heterogeneity and sparsity issues, enabling precise multi-label relation extraction. Experimental results show that the framework achieves a precision of 91.52%, a recall of 89.66%, and an F1-score of 90.58%, demonstrating its effectiveness and the synergistic contributions of its components.