A text sentiment classification model using double word embedding methods

Abstract

Sentiment analysis is an important topic in natural language processing (NLP) and text classifications. The existing algorithms of lexicon-based sentiment classification can deal with small corpus datasets or simple semantic texts. With the growth of text corpus data, word embedding methods have been gaining more attention. However, the single static word vector obtained by these methods can not accurately express the semantic information of the text. To optimize the word vector, we propose a text sentiment classification model using the double word embedding methods (DWE), which combines two models, GloVe and Word2vec, to represent the text to form a combinatory input of dual channels of convolution neural network (CNN). Based on the word vector fine-tuning strategy, the initial word vector is continuously learned and adjusted to find the CNN sentiment classification model with better combination input than a single vector representation. Experiment results show that DWE can effectively improve the accuracy of sentiment classification, which reaches 94.8%.