摘要
现有基于句法依存树的方面级情感分析模型大多只关注了句子的句法依赖结构,忽视了单词间的位置语义关系,同时现有模型只关注图卷积神经网络最后一层的输出,不能从不同的图卷积层学习。针对这个问题,提出了一种基于关系图卷积神经网络与双注意力的方面级情感分析模型。通过关系感知注意力抽取文本的位置语义关系,并与句法依存树结合,获取文本中丰富的结构信息,使用图卷积神经网络提取方面词的深层表示,使用双注意力机制融合不同图卷积层的输出,结合方面词的深层表示和上下文信息进行情感分类。在semval14和twitter数据集上的实验结果表明,与基准实验相比,关系图卷积网络和双注意力结构可以有效地提高模型的整体性能。
Most existing aspect-level sentiment analysis models based on syntactic dependency tree only focus on the syntactic dependency structure of sentences,ignoring the positional semantic relationship between words.At the same time,existing models only focus on the output of the last layer of the graph convolutional neural network and cannot learn from different graph convolutional layers.To address this problem,this paper proposes an aspect-level sentiment analysis model based on relational graph convolutional neural network and bidirectional attention.First,we extract the positional semantic relationship of the text through relationship-aware attention,and combine it with syntax dependency tree to obtain rich structural information in the text.Then we use graph convolutional neural network to extract the deep representation of aspect words.Finally,we use the bidirectional attention mechanism to fuse the output of different graph convolutional layers and combine the deep representation of aspect words and context information for emotional classification.Experimental results on semval14 and twitter datasets show that the graph convolutional network and the bidirectional attention structure can effectively improve the overall performance of the model compared with the benchmark experiment.
作者
方云龙
李卫疆
FANG Yunlong;LI Weijiang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,P.R.China;Key Laboratory of Artificial Intelligence of Yunnan Province,Kunming University of Science and Technology,Kunming 650500,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2023年第6期1164-1173,共10页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金项目(62066022)。
关键词
方面级情感分析
关系感知注意力
双注意力
图卷积神经网络
aspect-level sentiment analysis
relationship-aware attention
bidirectional attention
graph convolutional neural network
