基于改进胶囊网络的文本细粒度情感分类方法

Fine Grained Text Sentiment Classification Method Based on Improved Capsule Network

针对当前文本细粒度情感分类方法仅通过浅层卷积获取文本情感特征,导致多种文本细粒度情感分类效果差,具有歧义的文本细粒度情感分类精度低的问题,提出基于改进胶囊网络的文本细粒度情感分类方法。使用信息增益最大原则,优化文本特征集,引入文本特征词语位置信息,优化贝叶斯模型词语分辨性能,消除文本歧义。基于改进稠密胶囊网络模型,建立自注意力特征模型,提取文本细粒度情感特征,使用局部约束动态路由算法,选取与变换矩阵共享局部范围胶囊路由,实现文本细粒度情感分类。实验结果表明,所提方法的查准率、召回率以及F1值较高,多种文本细粒度情感分类效果较好,能够有效提高具有歧义的文本细粒度情感分类精度。

Aiming at the problem that the current text fine-grained emotion classification method only obtains the text emotion features through shallow convolution,resulting in the poor effect of various text fine-grained emotion classification and the low accuracy of ambiguous text fine-grained emotion classification,a text fine-grained emotion classification method based on improved capsule network is proposed.Using the principle of maximum information gain was used to optimize the text feature set,and the position information of text feature words was introduce to optimize the word resolution performance of Bayesian model,and eliminate text ambiguity.The feature model of self-attention was established via improving the dense capsule network model,and the fine-grained emotional features of the text were extracted.Local constrained dynamic routing algorithm was used to select and transform the matrix to share the local range of capsule routing,thus achieving fine-grained text sentiment classification.Experimental results show that the method has high precision,recall,F1 value,classification accuracy and excellent classification effect.