基于方面级情感分类的语义挖掘模型  

Semantic mining model based on aspect-level sentiment classification

作　　者：刘璐瑶 张换香[2] 张景 惠丽峰[2] Liu Luyao;Zhang Huanxiang;Zhang Jing;Hui Lifeng(Inner Mongolia University of Science and Technology,School of Information Engineering,Baotou,Inner Mongolia 014010,China;Inner Mongolia University of Science and Technology,School of Innovation and Entrepreneurship Education;Inner Mongolia University of Science and Technology,School of Science)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010 [2]内蒙古科技大学创新创业教育学院 [3]内蒙古科技大学理学院

出　　处：《计算机时代》2023年第6期65-68,75,共5页Computer Era

基　　金：国家自然科学基金项目(61562065);内蒙古自然科学基金项目(2019MS06001)。

摘　　要：方面级情感分类旨在确定句子中特定方面的情感极性。获取深层次方面级语义情感信息和方面级标记数据的缺乏是本领域研究的两个难点。本文提出一种基于语义注意力机制和胶囊网络的混合模型(SATTCap)。运用方面级归纳式迁移方式,将易获取的文档级评论知识中的情感语义迁移到方面级情感语义中,辅助方面级情感分类。另外基于重构语义依存的注意力机制提取深层次特征信息,采用方面路由方法,将深层次的方面级语义表示封装到语义胶囊中,然后采用Softmax预测。在公共数据集SemEval2014上对本文方法进行评估,结果表明,该模型在方面级情感分类任务上的表现是有效的。Aspect-level sentiment classification aims to determine the polarity of a particular aspect of a sentence.Obtaining deep aspect-level semantic sentiment information and the lack of aspect-level labeled data are two difficulties in this field.In this paper,a hybrid model based on semantic attention mechanism and capsule network(SATTCap)is proposed.The aspect-level inductive transfer method is used to transfer the sentiment semantics from easily accessible document-level comment knowledge to aspectlevel sentiment semantics,which assists aspect-level sentiment classification.In addition,the deep feature information is extracted based on the attention mechanism of reconstructed semantic dependency.The deep aspect-level semantic representation is encapsulated into the semantic capsule by the aspect routing method,and then Softmax is used for sentiment prediction.The proposed method is evaluated on the public dataset SemEval 2014.Experimental results show that it is effective in the performance of aspect-level sentiment classification task.

关 键 词：方面级 情感分类 注意力机制 胶囊网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]