摘要
针对当前神经网络模型未充分考虑标注词语之间的语义关联和预测标签之间的约束关系的问题,提出了一种基于历史信息的目标属性抽取模型。该模型使用注意力机制融合历史的语义信息和预测标签的约束关系,增强特征信息的抽取能力,并利用目标属性和观点项的共现现象进一步提高标注的准确率。实验结果表明,该模型在一定程度上能够提高目标属性抽取网络的性能。
Aiming at the current neural network model of semantic relationship between words and the constraint relationship between labels are not fully considered.A target attribute extraction model combining historical information is proposed.The model can enhance the extraction ability of feature information by using the attention mechanism to fuse the historical semantic information and the constraint relationship of the prediction tag,and can further improve the accuracy of the annotation by using the co-occurrence phenomenon of the target and opinion term.Experiments show that the model can improve the performance of the target attribute extraction network to a certain extent.
作者
彭军
韦照川
PENG Jun;WEI Zhaochuan(School of Information and Communication Engineering,Guilin University of Electronic Technology,Guilin 541004,China)
出处
《桂林电子科技大学学报》
2020年第3期183-188,共6页
Journal of Guilin University of Electronic Technology
基金
广西无线宽带通信与信号处理重点实验室基金(GXKL0614107)
新疆重点研发计划(2018B03022-2)。
关键词
目标属性抽取
历史信息
序列标注
长短时记忆网络
aspect term extraction
historical information
sequence labeling
long-short term memory network
作者简介
通信作者:韦照川(1973-),男,副教授,研究方向为信号处理、北斗导航与定位。E-mail:149453501@qq.com。
