关键词: 关系抽取, 掩码, 门控神经网络, 预训练语言模型, 提示学习

Abstract: Abstract: To tackle the difficulty of semantic mining of entity relations and biased relation prediction in relation extraction tasks, a relation extraction method based on mask and gated memory network calibration was proposed. First of all, the latent semantics between entities within the pre-training language model semantic space were effectively learned through the utilization of masks in s. By constructing a mask attention weight matrix, the discrete masked semantic spaces were interconnected. And then, the gate-controlled calibration networks were incorporated to integrate the masked representations, which contain entity and relation semantics. Besides, these calibrated representations served as s to adjust the relation information, and the final representation of the calibrated sentence was mapped to the corresponding relation class. Finally, the potential of pre-training language models was fully exploited by the proposed approach through harnessing the power of masks in s and combining it with the advantages of traditional fine-tuning methods.The experimental results highlight the effectiveness of the proposed method. On the SemEval (SemEval-2010 Task 8) dataset, the F1 score reaches an impressive 91.4%, outperforming the RELA (Relation Extraction with Label Augmentation) generative method by 1 percentage point and reducing the parameter count by approximately 10% compared to RELA. Additionally, the extraction performances on the SciERC (Entities, Relations, and Coreference for Scientific) and CLTC (Chinese Literature Text Corpus) datasets are remarkable, achieving 91.0% and 82.8% respectively. The effectiveness of the proposed method is evident as it consistently outperforms the comparative methods on all three datasets mentioned above, establishing itself as the current state-of-the-art approach. Furthermore, superior extraction performance is achieved by this method compared to generative methods, while utilizing fewer parameters.

Key words: relation extraction, mask, gated neural network, pre-trained language model, tuning