Sentiment Parsing of Chinese Microblogs Using Recurrent Neural Network  

作　　者：程佳军 张鑫 张胜 王晖 刘博 

机构地区：[1]College of Information Systems and Management, National University of Defense Technology, Changsha 410073, China [2]Air Force Electromagnetic Spectrum Management Center, Beijing 100000, China

出　　处：《Journal of Donghua University(English Edition)》2016年第3期501-504,共4页东华大学学报（英文版）

基　　金：National Natural Science Foundation of China(No.71331008)

摘　　要：Easy accessibility and light content filtering attempt have made microblogging sites the most popular platforms for users to share their experiences and express their opinions.Extracting from the user-composed microblogs the opinions expressed are of great significance for many practical applications.However,such task is very challenging,in particular for Chinese Microblogs.A novel representation of the opinions expressed in microblog sentences is presented and a recurrent neural network(RNN) based sequence labeling approach is proposed about sentiment parsing of Chinese microblogs.The experiments evaluate the performance of different RNN models and explore the bi-directional and deep versions of each model on a Chinese microblog corpus built by this paper.Experimental results show that the bidirectional version of the gated recurrent unit(GRU) model with three layers achieves the highest F-score 0.622.Easy accessibility and light content filtering attempt have made microblogging sites the most popular platforms for users to share their experiences and express their opinions.Extracting from the user-composed microblogs the opinions expressed are of great significance for many practical applications.However,such task is very challenging,in particular for Chinese Microblogs.A novel representation of the opinions expressed in microblog sentences is presented and a recurrent neural network（RNN） based sequence labeling approach is proposed about sentiment parsing of Chinese microblogs.The experiments evaluate the performance of different RNN models and explore the bi-directional and deep versions of each model on a Chinese microblog corpus built by this paper.Experimental results show that the bidirectional version of the gated recurrent unit（GRU） model with three layers achieves the highest F-score 0.622.

关 键 词：sentiment analysis recurrent nueral network Chinese Micriblog 

分 类 号：TH17[机械工程—机械制造及自动化]