摘要
方面级情感分析是如今情感分析领域的重要研究任务之一,旨在计算文本中多个方面词的情感极性。现有的方面级情感分析方法通常将整个句子直接输入复杂的神经网络,尽管此类方法能够有效捕捉到词与词之间的依赖关系,却忽略了方面词与其上下文之间所隐含的位置特征。因此,提出了一种融合位置特征的方面级情感分析方法。将文本分别采用基于方面词间隔的上下文划分方式与基于单词距离的上下文划分方式,通过两个微调后的BERT模型,完成词向量的表达;将两种词向量送入多头注意力机制,计算其文本特征;使用平均池化将语义信息进行融合,在输出层完成方面词的情感极性分类。在SemEval2014 Task4数据集和Twitter数据集上的实验表明,提出的融合位置特征的方面级情感分析方法能够充分利用方面词上下文之间的位置特征,有效提升了准确率和F1值。
Aspect-level sentiment analysis is one of the important research tasks in the field of sentiment analysis,aiming at calculating the sentiment polarity of various aspect words in the text.The existing aspect-level sentiment analysis methods usually input the whole sentence directly into complex neural networks.Although this kind of method can effectively capture the dependency between words,it ignores the implicit location features between aspect words and its contexts.Therefore,we propose an aspect-level sentiment analysis method with location features.The aspect words and their contexts are divided into context based on aspect word interval and context based on word distance,respectively,and the expression of word vectors is completed by two fine-tuned BERT models.Two kinds of word vectors are sent into the multi-head attention mechanism,and their text features are calculated.Average pooling is used to fuse semantic information,and the emotional polarity classification of aspect words is completed at the output level.Experiments on SemEval2014 Task4 data set and Twitter data set show that the proposed aspect-level sentiment analysis method with location features can make full use of the location features between the contexts of aspect words,and effectively improve the accuracy and F1-Measure.
作者
翟社平
成大宝
张文晴
刘园彪
ZHAI She-ping;CHENG Da-bao;ZHANG Wen-qing;LIU Yuan-biao(School of Computer Science&Technology,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)
出处
《计算机技术与发展》
2023年第5期167-172,共6页
Computer Technology and Development
基金
工业和信息化部通信软科学项目(2018R26)
陕西省重点研发计划项目(2022GY-038)
国家级大学生创新创业训练计划项目(202111664004)。
关键词
方面级情感分析
位置特征
注意力机制
BERT
深度学习
aspect-level sentiment analysis
location features
attention mechanism
BERT
deep learning
