MSAM: Self-Attention-Based Joint Extraction of Entities and Relations from Multi-Dimensional Span Information and Its Application in Power Emergency Systems

With the increasing complexity of power systems, their stability and reliability are crucial to the economic development and social operation of modern society. Improving the efficiency of emergency response to power failures, accurately extracting useful information from massive text data, and effectively organizing such information have become key research focuses. Existing methods still face limitations in handling long-distance dependencies in long texts and context understanding, leading to error propagation and insufficient semantic comprehension. To address these issues, this paper proposes a Multi-dimensional Span-based Attention Mechanism (MSAM) for the joint extraction of entities and relations. By constructing a unified model, MSAM simultaneously performs named entity recognition and relation extraction tasks, effectively capturing both long and short-distance dependencies in text, enhancing context understanding, and reducing error propagation. Compared with state-of-the-art models, experimental results on two public datasets demonstrate that MSAM achieves significant improvements, particularly in relation extraction. Furthermore, the application of MSAM on a self-constructed power emergency dataset verifies its effectiveness in practical scenarios.