【摘要】 为解决传统方法在短文本实体链接中因特征不足导致的准确率下降问题，提出一种基于实体间提示的迭代式实体链接方法。该方法使用预训练模型对实体提及进行分类，提高候选实体生成的准确性；通过注意力机制识别上下文中的关键关联实体，并结合其知识库描述生成链接提示；为了避免初期提示造成的噪音和错误问题，对目标实体提及进行多轮迭代式链接，逐步提高链接准确率。在两个公开数据集上的实验结果显示，该方法在短文本实体链接任务上优于现有技术，尤其在小样本情况下表现出色，可有效应对上下文信息不足和样本量有限的挑战。更多还原

【Abstract】 To address the accuracy degradation in short-text entity linking caused by insufficient features in traditional methods, an iterative entity linking approach based on inter-entity prompts was proposed. The method first employs a pre-trained model to classify entity mentions, thereby improving candidate entity generation accuracy. The entity mentions were classified using a pretrained model, thereby improving candidate entity generation accuracy. Key contextual entities were identified through an attention mechanism, and linking prompts were generated by combining their knowledge base descriptions. To mitigate noise and errors introduced by initial prompts, multi-round iterative linking was performed on the target entity mention, progressively enhancing the linking accuracy. Experimental results on two public datasets demonstrate that the proposed method outperforms existing techniques in short-text entity linking tasks, particularly showing superior performance in few-shot scenarios. The approach effectively addresses challenges posed by insufficient contextual information and limited sample sizes.更多还原