基于神经网络的方面级情感分析模型  

Aspect-Based Sentiment Analysis Model Based on Neural Network

作　　者：马晓宁[1] 赵志峰 MA Xiao-ning;ZHAO Zhi-feng(College of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300

出　　处：《计算机仿真》2022年第11期491-495,共5页Computer Simulation

基　　金：中央高校基本科研业务费中国民航大学专项基金项目(3122014C018)。

摘　　要：在线评论中包含用户的情感倾向和感情态度,而且一条评论往往包含用户对商家多个方面的评价态度。针对评论的方面级情感分析问题,提出一种融合层次注意力机制和多任务学习的Multitask-Attention-BiGRU神经网络模型。模型在特征提取阶段共享参数,建立多任务学习模型,并在输出层做区分和单独输出,可对用户评价进行多方面情感分析。模型引入注意力机制分别在词语级别、句子级别以及篇章级别进行特征提取,提高神经网络信息提取效率。通过实验表明,在评论方面级情感分析问题中,相较于其它对比模型,Multitask-Attention-BiGRU模型的精确度和F1值更好,并且效率更高。Generally,sentiment tendency and sentiment attitude belong to the category of online comments,and a comment contains the user’s evaluation of the business in many aspects.In order to solve the aspect level sentiment analysis of reviews,this paper proposes a Multitask-Attention-BIGRU neural network model which integrates hierarchical attention mechanism and multitask learning.In the process of feature extraction,the data of the model were shared to build a multi task learning model,and the output layer was used to distinguish and output separately,so that users can evaluate and analyze their emotions in various aspects.Attention mechanism was introduced into the model to extract features at word level,sentence level and text level,thus improving the efficiency of neural network information extraction.The experimental results show that compared with other comparison models,the Multitask-Attention-BiGRU model has better accuracy and F1 value,and is more efficient.

关 键 词：注意力机制 多任务学习 方面级情感分析 神经网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]