基于位置增强和双向图卷积的方面级情感分析  

Aspect-Based Sentiment Analysis Based on Position Enhancement and Bidirectional Graph Convolution

作　　者：郭钰铉 韩萌 李晖[1] GUO Yuxuan;HAN Meng;LI Hui(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870

出　　处：《微处理机》2023年第4期38-41,共4页Microprocessors

摘　　要：方面级情感分析旨在判别给定句子中具体方面的情感极性。针对大多数模型对位置信息关注度不够,以及单向图卷积神经网络不能充分应用单词间的依存关系较好地识别方面词的情感倾向等问题,提出基于位置增强和双向图卷积的方面级情感分析模型。模型将位置信息和上下文词向量进行融合,并通过双向长短时记忆网络捕获语义信息;利用双向图卷积神经网络为提取方面特征提供句法约束,并通过掩码层得到特定的方面特征;通过注意力机制学习上下文与特定方面之间的重要信息。与ASGCN模型相比,该模型在Rest14、Lap14和Twitter三个公开数据集上的准确率和F1值都有提升。Aspect-based sentiment analysis aims to identify the emotional polarity of specific aspects in a given sentence.Aiming at the problems that most models don't pay enough attention to position infor-mation,and the one-way graph convolution neural network can't fully apply the dependency between words to identify the emotional tendency of aspect words,an aspect-based sentiment analysis model based on location enhancement and bidirectional graph convolution is proposed.The model fuses position infor-mation and context word vectors,and captures semantic information through bidirectional long short-term memory networks.Bidirectional graph convolution neural network is used to provide syntactic constraints for extracting aspect features,and specific aspect features are obtained through mask layer.Important information between context and specific aspects is learned through attention mechanism.Com-pared with ASGCN model,the accuracy and F1 value of the model are improved on the three public data sets of Rest14,Lap14 and Twitter.

关 键 词：方面级情感分析 位置信息 双向图卷积网络 注意力机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]