摘要
针对目前用于文本情感分析神经网络非常缺乏的问题,提出了一种级联RNN的体系结构。该体系结构首先将RNN放在全局平均池化层上,用于捕获与CNN之间的长期依赖关系,然后通过GloVe嵌入方法对词向量进行处理,最终作为输入数据,进行训练。该方法与Twitter语料库中的基线模型相比,实验表现出更好的情感分类效果,该方法在Twitter情感语料库中最高识别率可达88.86%,从而为情感分析提供可靠的依据。并且它具有超参数调整功能,能够减少更高性能的参数的数量。
Aiming at the problem of lacking neural network for text sentiment analysis,a cascade RNN architecture is proposed.In this architecture,RNN is firstly placed on the global average pooling layer to capture the long-term dependency relationship with CNN,and then word vectors are processed by GloVe embedding method and finally used as input data for training.Compared with the baseline model in the Twitter corpus,the method shows a better effect of sentiment classification.The highest recognition rate of this method in the Twitter sentiment corpus is 88.86%,thus providing reliable basis for sentiment analysis.And it has a hyperparameter adjustment function,which can reduce the number of parameters with higher performance.
作者
郭勇
赵康
潘力
GUO Yong;ZHAO Kang;PAN Li(North Sichuan College of Preschool Teacher Education,Guangyuan 628000,Sichuan Province,China;Shangqiu Vocational and Technical College,Shangqiu 476100,Henan Province,China;Zhengzhou University of Technology,Zhengzhou 450044,China)
出处
《信息技术》
2021年第2期50-55,共6页
Information Technology
基金
河南省科技厅科技攻关计划项目(202002210346)。
