多维特征融合的混合神经网络文本情感分析模型

Multi-dimensional Feature Fusion Hybrid Neural Network Text Sentiment Analysis Model

针对复杂文本评论的情感分析研究存在着隐式主题方面分类不精确、文本特征提取不全面和识别文本上下文语义不足等问题,论文提出了一种多维特征融合的混合神经网络文本情感分析模型MFF-HNN.该模型先把词向量、词性、位置和句法依存特征进行注意力特征融合,抽取出主题词库,然后把融合特征输入到改进的TBGRU模型和DCNN模型中获取语义信息和局部特征信息,再与主题词库结合进行注意力特征融合语义特征信息,最后使用SoftMax函数获取文本方面级情感分类信息.实验表明,该模型的情感分类的效果优于其它模型.

Aiming at the problems of inaccurate classification of implicit topics,incomplete text feature extraction,and insufficient semantic recognition of text context in the sentiment analysis research of complex text reviews,this paper proposes a multi-dimensional feature fusion hybrid neural network text sentiment analysis model MFF-HNN.The model first fuses the attention feature of word vector,part of speech,position and syntactic dependency,extracts the thesaurus,and then inputs the fused features into the improved TBGRU model and DCNN model to obtain semantic information and local feature information,and then combines with The thesaurus is combined with the attention feature to fuse the semantic feature information,and finally the SoftMax function is used to obtain the text aspect-level sentiment classification information.Experiments show that the emotion classification effect of this model is better than other models.