基于双特征嵌套注意力的方面词情感分析算法  

An algorithm for aspect-based sentiment analysis based on dual features attention-over-attention

作　　者：肖宇晗 林慧苹[1] 汪权彬 谭营[2] XIAO Yuhan;LIN Huiping;WANG Quanbin;TAN Ying(School of Software and Microelectronics,Peking University,Beijing 102600,China;School of Electronics Engineering and Computer Science,Peking University,Beijing 100871,China)

机构地区：[1]北京大学软件与微电子学院,北京102600 [2]北京大学信息科学技术学院,北京100871

出　　处：《智能系统学报》2021年第1期142-151,共10页CAAI Transactions on Intelligent Systems

基　　金：国家重点研发计划项目(2018AAA0102301,2018AAA0100302,2018YFB1702900);国家自然科学基金项目(62076010)。

摘　　要：针对目前方面词情感分析方法忽视了以方面词为核心的局部特征的重要性,并难以有效减小情感干扰项的负面噪声的问题,本文提出了一种带有基于变换器的双向编码器表示技术(bi-directional encoder representations from transformers,BERT)加持的双特征嵌套注意力模型(dual features attention-over-attention with BERT,DFAOA-BERT),首次将AOA(attention-over-attention)与BERT预训练模型结合,并设计了全局与局部特征提取器,能够充分捕捉方面词和语境的有效语义关联。实验结果表明:DFAOA-BERT在SemEval 2014任务4中的餐馆评论、笔记本评论和ACL-14 Twitter社交评论这3个公开数据集上均表现优异,而子模块的有效性实验,也充分证明了DFAOA-BERT各个部分的设计合理性。Aspect-based sentiment analysis is of great significance to making full use of product reviews to analyze potential user needs.The current research work still has deficiencies.Many studies ignore the importance of local features centered on aspects and fail to handle emotional disturbances effectively.To address these problems,this article proposes a dual features attention-over-attention model with BERT(DFAOA-BERT).For the first time,an AOA(attentionover-attention)mechanism is combined with the BERT pretrained model.DFAOA-BERT also designs global and local feature extractors to fully capture an effective semantic association between aspects and context.According to the experimental results,DFAOA-BERT performs well on three public datasets:restaurant and laptop review datasets from SemEval 2014 Task 4 and the ACL-14 Twitter social review dataset.The effectiveness experiment of submodules also fully proves that each part of DFAOA-BERT makes a significant contribution to the excellent performance.

关 键 词：情感分析 方面词 嵌套注意力 BERT预训练模型 全局特征 局部特征 深度学习 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]