摘要
在实体关系抽取和分类的任务背景下,针对当前基于深度学习模型对实体间语义关系利用不足的问题,提出了基于预训练ALBERT模型的实体关系分类模型。首先使用实体定位符号对实体位置进行定位,然后使用依存关系方法提取实体间关系描述词,并利用预训练ALBERT模型对输入进行向量化表征,再利用注意力机制将向量进行融合,最终使用Softmax分类器得到关系分类结果。与其他几种深度模型相比,该模型在KBP37数据集和Semeval-2010-task8数据集上的结果最优,验证了模型的有效性。
In the context of entity relation extraction and classification tasks,a relation extraction model based on pre-training ALBERT model is proposed to solve the problem of insufficient utilization of semantic relationship between entities based on the deep learning model.Entity positioning symbols are used to locate entity positions,the dependency method is used to extract the relationship indicator pronouns between entities,and then the pre-training ALBERT model is used to embed the input.Then the attention mechanism is used to fuse the learning weights of the vectors,and relationship extraction results are obtained using the Softmax classifier.Compared with several other deep models,this model has the best results on the KBP37 dataset and Semeval-2010-task8 dataset,which verifies the effectiveness of the model.
作者
朱瑞天
王兴芬
ZHU Ruitian;WANG Xingfen(Computer School,Beijing Information Science&Technology University,Beijing 100101,China;School of Information Management,Beijing Information Science&Technology University,Beijing 100192,China)
出处
《北京信息科技大学学报（自然科学版）》
2021年第5期20-27,共8页
Journal of Beijing Information Science and Technology University
基金
国家重点研发计划资助项目(2019YFB1405003)。
