A Feature-Enhanced Joint Relation Extraction Neural Network Based on State Space Models

The joint extraction of entities and relationships from unstructured text is a critical task in information extraction. Despite recent advancements, challenges such as multi-relational triplet extraction, long-distance dependencies, and similar relations remain. This paper proposes a feature-enhanced joint relation extraction neural network based on state-space models, designed to improve the accuracy and efficiency of relation extraction. The model integrates the Mamba2 module with a parallel multi-head attention mechanism to capture textual semantics and uses an enhancement module to address data scarcity and fine-grained redundancy. Experimental results show that the proposed framework achieves state-of-the-art performance on the HacRED dataset, with an F1 score of 64.5, demonstrating its effectiveness in multi-relational extraction and distinguishing semantically similar relations.