基于贝叶斯网的跨领域情感分析方法  

CROSS-DOMAIN SENTIMENT ANALYSIS BASED ON BAYESIAN NETWORK

作　　者：刘慧清 郭延哺 李维华[1] Liu Huiqing;Guo Yanbu;Li Weihua(School of Information Science and Engineering,Yunnan University,Kunming 650500,Yunnan,China)

机构地区：[1]云南大学信息学院,云南昆明650500

出　　处：《计算机应用与软件》2020年第12期119-126,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61762090);云南省应用基础研究计划重点项目(2016FA026);云南省创新团队项目(2018HC019);云南大学研究生科研创新基金项目(2018226)。

摘　　要：基于机器学习的情感分析依赖于充足的标签样本。针对标签样本不足以及情感分类器存在的领域适应性问题,提出一种基于贝叶斯网的跨领域情感分析方法。基于贝叶斯网,对源领域和目标领域构建局部特征模型;研究局部特征模型的融合方法并构建全局特征模型;基于全局特征模型建立情感知识的迁移方法并训练分类器。在Amazon数据集上进行实验,结果表明,该方法在一定程度上提高了目标领域文本的情感分类精度,以及分类器在目标领域中的适应性。Sentiment analysis based on machine learning relies on ample labeled data.Aiming at the problem of insufficient label data and the domain adaptation of sentiment classifiers,we propose a cross-domain sentiment analysis based Bayesian network.Based on the Bayesian network,the local feature models were constructed for the source domain and the target domain;the fusion method of the local feature models was studied and the global feature model was established;the sentiment feature transfer method was established based on the global feature model and classifiers were trained.Finally,the experiments were carried out on Amazon dataset.The experimental results show that our method improves the sentiment classification accuracy of the target domain texts and the adaptation of the classifier in the target domain to some extent.

关 键 词：跨领域情感分析 贝叶斯网 融合 迁移 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]