Sentiment Analysis for E-Commerce Reviews Based on Deep Learning Hybrid Model.

Along with the emergence of Internet, online shopping has become a fundamental way for consumers to purchase and consume. Sentiment analysis of huge reviews on related e-commerce platforms can effectively improve user satisfaction. In this paper, a two-channel hybrid model sentiment classification algorithm MWVH is proposed, which combines the features of multiple word vectors. For words, word vectors are extracted by combining Word2vec and character-level N-gram embedding methods on different parallel channels. CNN and RNN combined with attention mechanism are each on a channel. Two-channel model structure are adopted to better obtain sentence vector features, which makes up for the deficiency of single network model. Using the data set of 8 cross-border e-commerce APP reviews from APP Store, comparative experiments were conducted on the 2 models (MWVH1.0 and MWVH2.0) and 4 benchmark models proposed. Experimental results show that the proposed two-channel hybrid model (MWVH1.0 and MWVH 2.0) based on the embedding features of multiple words can improve the classification performance. When the amount of data is small, MWVH1.0 model is better, with an average accuracy of 94.62%. However, MWVH2.0 performs better on larger data sets, with an average accuracy of 95.14%.