基于迁移学习和过滤机制的方面级情感分析

TLFM:Aspect-level sentiment analysis based on transfer learning and filtering mechanism

考虑到迁移学习过程中引入大量与给定方面无关的情感噪音,提出一种基于迁移学习和过滤机制的方面级情感分析模型TLFM。利用预训练文档级情感分析模块学习文档的情感知识,通过共享参数的方式将情感知识传递给方面级情感分析模块;设计一个注意力过滤模块,该模块聚焦于过滤文档级知识中与给定方面无关的情感;将预学习后的模型TLFM和过滤模块进行联合训练,利用文档级知识的同时,降低噪音的影响。实验结果表明,迁移学习和过滤机制的结合能有效提高方面级情感预测的准确率。

Considering that the transfer learning process causes a large number of emotional noise unrelated to the given aspect,an aspect-level sentiment analysis model TLFM based on transfer learning and filtering mechanism was proposed.The pre-trai-ning document-level sentiment analysis module was used to learn the emotional knowledge of the document,and the emotional knowledge was transmitted to the aspect-level sentiment analysis module by sharing the parameters.An attention filtering module was designed,which focused on filtering emotions unrelated to the given aspects in document-level knowledge.The pre-learning model TLFM and filtering module were jointly trained to reduce the impact of noise while using document level know-ledge.Experimental results show that the combination of transfer learning and filtering mechanism can effectively improve the accuracy of aspect level emotion prediction.