Abstract
Sentiment classification, as a core part in Natural Language Processing, has achieved great improvements in recent years. With the development of the Internet, online social platforms have become increasingly more popular, attracting people to publish numerous user-generated contents that contain important voice and sentiment information. Existing sentiment classification works are mostly based on the original text, while these texts always suffer a lot from typos, some of which even change the original meaning of sentence. In this paper, a Spelling Check based Neural Network for Sentiment Classification (SCNNSC) is purposed. First, a spelling check method is used to conduct text correction in advance, in order to avoid being influenced by common spelling or typing error. Then, we introduce CNN-GRU-Cross (CGC) model for sentiment classification. By utilizing multiple word embedding learning algorithms, and the cross structure of CNN and BiGRU, our model can extract both local semantic features as well as long term dependency feature, thus capturing more comprehensive sentence representation. Finally, by integrating spelling check method and CGC model, SCNNSC can be obtained. Experiments are conducted on Chinese Microblog dataset and Taobao dataset. Experimental results indicate that our SCNNSC can achieve satisfying performance on sentiment classification.