句法感知与知识增强的方面级情感分析  

Aspect-based sentiment analysis of syntactic perception and knowledge enhancement

作　　者：陈可嘉 张雨鹏 林鸿熙 CHEN Kejia;ZHANG Yupeng;LIN Hongxi(School of Economics and Management,Fuzhou University,Fuzhou 350108,China;School of Business,Putian University,Putian 351100,China)

机构地区：[1]福州大学经济与管理学院,福建福州350108 [2]莆田学院商学院,福建莆田351100

出　　处：《西安电子科技大学学报》2024年第5期165-178,共14页Journal of Xidian University

基　　金：国家自然科学基金(72374030);福建省财政厅专项资金(0300-83022110)。

摘　　要：在方面级情感分析任务中需要先对文本进行句法依存解析,一方面存在着高度依赖依存解析质量的问题,另一方面未考虑依存解析与语义知识关联性缺失的问题。为此,提出一种基于句法感知与知识增强的双通道图卷积模型进行方面级情感分析任务。在一个通道上使用句法感知机制学习句子的依存关系,在另一个通道上通过知识图谱进行知识增强,并通过信息交互机制关联两通道的输出,使得模型对与方面词相关联的重要词汇赋予更多句法与语义上的关注。此外,引入位置注意力机制,对词语进行位置上的得分权重调整,进而提高方面级情感分析任务的性能。在3个公开数据集Rest14、Lap14、Twitter上进行实验,相较于其他方面级情感分析模型,所提模型在准确率与F1值都有较为明显的提升。实验表明,句法感知与知识增强能够指导图卷积模型进行更加深入的语义学习与合理的权重分配,从而提高方面级情感分析任务的性能。In the aspect-based sentiment analysis task,the syntactic dependency parsing of the text is required first,which is highly dependent on the quality of the dependency parsing and does not take into account the lack of correlation between dependency parsing and semantic knowledge.Therefore,a two-channel graph convolution model based on syntactic perception and knowledge enhancement is proposed for the aspect-based sentiment analysis task.Syntax-perception mechanisms are used to learn sentence dependencies in one channel,and knowledge enhancement is performed in the other channel through a knowledge graph,with the outputs of the two channels correlated through an information interaction mechanism,which allows the model to pay more syntactic and semantic attention to important words associated with aspectual words.In addition,a positional attention mechanism is introduced to adjust the score weights of words with respect to the position,which in turn improves the performance of the aspect-based sentiment analysis task.Experiments are conducted on three public datasets,Rest14,Lap14 and Twitter.Compared to other aspect-based sentiment analysis models,this paper’s model shows a more significant improvement in both accuracy and F1 value.Experiments show that syntactic perception and knowledge enhancement can guide the graph convolutional model to perform deeper semantic learning and reasonable weight allocation,thus improving the performance of aspect-based sentiment analysis tasks.

关 键 词：情感分析 图卷积神经网络 依存关系 知识图谱 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]