融合关键对象识别与深层自注意力的Bi-LSTM情感分析模型

Sentiment Analysis Model of Bi-LSTM with Key Opinion Target Recognition and Deeper Selfattention

在线评论文本通常涉及多个评价对象,对象的表达方式有显式和隐式之分,针对不同对象的情感倾向可能不会完全一致.关键评价对象是评论中最受关注的对象,其相应的情感语义对整条评论的情感观点起主导作用.本文构建了融合关键对象识别与深层自注意力机制的Bi-LSTM模型,以提升短文本情感分类的效果.使用CNN处理文本,基于卷积层输出结果识别关键评价对象,并在此基础上完成深层自注意力的学习.将对象信息与文本信息进行融合,利用注意力机制强化的Bi-LSTM模型得到评论文本的情感分类结果.在酒店评论数据集上进行实验,与之前基于深度学习的模型相比,本文方法在精确率、召回率和F-score评价指标方面均有更好的表现.

Online comment texts usually involve multiple opinion targets,which can be expressed in explicit and implicit ways,while the emotional tendencies of different targets may not be exactly the same.The key opinion target is the most concerned aspect in the comment,and its corresponding emotional semantics plays a leading role in the emotional view of the whole comment.In order to improve the effect of short text emotion classification,this paper constructs Bi-LSTM model which integrates key opinion target recognition and deeper self-attention mechanism.First,CNN is used to process the text,and the key opinion target is identified based on the output of convolution layer.On these grounds,deep self-attention learning is accomplished.Then,key opinion target information and text information are fused,and the Bi-LSTM model strengthened by attention mechanism is used to get the emotion classification results of comment texts.A comparative experiment is conducted on the evaluation dataset.Compared with the previous deep learning based models,the proposed method in this paper has better performance in evaluation indicators such as accuracy rate,recall rate and F-score.