归纳式迁移学习在跨领域情感倾向性分析中的应用

Application of inductive transfer learning in cross-domain sentiment analysis

在解决情感倾向性分析问题中,传统的监督学习和半监督学习都是在训练和测试所用的数据来自相同分布的假设基础之上的,但在很多情况下不能满足这样的假设,这就产生了跨领域的情感倾向性分析问题.在跨领域情感倾向性分析中,提出一种基于归纳式迁移学习的图模型,通过图模型建立源领域和目标领域数据之间的关联,使得源领域的数据通过图模型学习目标领域数据在特征和实例上的特点.同时,利用归纳式迁移学习方法使用少量的目标领域的已标注数据进行训练,从而提高了情感分类器在目标领域的分类准确率,极大地改进了跨领域情感倾向性分析的效果.在标准数据集上进行了实验,并与监督学习方法 SVM、半监督学习方向TSVM以及其它3种常用的迁移学习方法进行了对比,对比结果显示本文方法显著的高于SVM和TSVM,并在大多数数据集上优于其它3种迁移学习方法,实验结果表明该方法是有效的.

To solve the sentiment analysis problems,traditional supervised learning methods and semi-supervised learning methods are based on the assumption that the training data and the testing data come from the identical distribution.It generates the cross-domain sentiment analysis problems when the assumption was not satisfied in some cases.In this paper,aiming at the problems of cross-domain sentiment analysis,agraph model is proposed based on the inductive transfer learning.The relevance of the source domain and the target domain data was built,such that the source domain data could learn the characteristic at the point of the feature and the instance of the target domain data by the graph model.Simultaneously,the accuracy of sentiment classifiers in the target domain could be improved by using few target domain labeled data by the inductive transfer learning method.The result of cross-domin sentiment analysis could be significantly improved.Experiments were carried out on the standard data,and were compared with supervised learning approach SVM,semi-supervised learing approach TSVM,as well as other three state-of-the-art transfer learning approaches.The results show that the proposed approach is superior to SVM and TSVM significantly,and is superior to other three state-of-the-art transfer learing approaches at almost every detaset.The results show the efficiency of the proposed approach.