Title: A topic-enhanced recurrent autoencoder model for sentiment analysis of short texts

Authors: Shaochun Wu; Ming Gao; Qifeng Xiao; Guobing Zou

Addresses: Department of Intelligent Information Processing, Shanghai University, Shanghai 200444, China ' Department of Intelligent Information Processing, Shanghai University, Shanghai 200444, China ' Department of Intelligent Information Processing, Shanghai University, Shanghai 200444, China ' Department of Intelligent Information Processing, Shanghai University, Shanghai 200444, China

Abstract: This paper presents a topic-enhanced recurrent autoencoder model to improve the accuracy of sentiment classification of short texts. First, the concept of recurrent autoencoder is proposed to tackle the problems in recursive autoencoder including 'increasing in computation complexity' and 'ignoring the natural word order'. Then, the recurrent autoencoder model is enhanced with the topic and sentiment information generated by joint sentiment-topic (JST) model. Besides, in order to identify the negations and ironies in short texts, sentiment lexicon is utilised to add feature dimensions for sentence representations. Experiments are performed to determine the feasibility and effectiveness of the model. Compared with recursive autoencoder model, the classification accuracy of our model is improved by about 7.7%.

Keywords: short texts; sentiment analysis; recurrent autoencoder; recurrent neural network; joint sentiment-topic model.

DOI: 10.1504/IJIMS.2020.110230

International Journal of Internet Manufacturing and Services, 2020 Vol.7 No.4, pp.393 - 406

Received: 14 Jul 2018
Accepted: 15 Nov 2018

Published online: 18 Apr 2020 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article