A Hierarchical multi-input and output Bi-GRU Model for Sentiment Analysis on Customer Reviews

2017 International Symposium on Application of Materials Science and Energy Materials

材料科学;能源学

Zhang, Liujie^1 ; Zhou, Yanquan^1,2 ; Duan, Xiuyu^1 ; Chen, Ruiqi^1

School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China^1

Engineering Research Center of Information Networks, Ministry of Education, Beijing, China^2

关键词: Bi-directional;    Computationally efficient;    Customer review;    Emotional expressions;    Lexical information;    Model-based OPC;    Part Of Speech;    Sentiment classification;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/322/6/062007/pdf
DOI  :  10.1088/1757-899X/322/6/062007

学科分类：材料科学（综合）

来源: IOP

【 摘 要 】

Multi-label sentiment classification on customer reviews is a practical challenging task in Natural Language Processing. In this paper, we propose a hierarchical multi-input and output model based bi-directional recurrent neural network, which both considers the semantic and lexical information of emotional expression. Our model applies two independent Bi-GRU layer to generate part of speech and sentence representation. Then the lexical information is considered via attention over output of softmax activation on part of speech representation. In addition, we combine probability of auxiliary labels as feature with hidden layer to capturing crucial correlation between output labels. The experimental result shows that our model is computationally efficient and achieves breakthrough improvements on customer reviews dataset.