Support vector machine equipped with deep convolutional features for product reviews classification

Online shopping websites usually ask their customers to review the products and associated services. As e-commerce is becoming more and more popular, the number of customer reviews grows rapidly. It is difficult for a potential customer to read all original reviews in order to make a decision on whether to buy the product. So product reviews classification has grown to be one of the hottest research areas in sentiment analysis. In practice, sentiment analysis usually has two methods: Support Vector Machine (SVM) and Convolutional Neural Network (CNN). The former uses one-hot representation as features with simple network and competitive results. The latter uses distributed representation or word embedding as features. SVM with one-hot representation is fast but may not extract suitable features of specific texts. CNN with distributed representation can get impressive results but usually on the basis of an extremely huge and complex network. In this paper, we proposed an original architecture called CNN-SVM for product reviews classification. SVM classifier equipped with deep convolutional features is utilized, which combines the terrific discriminative capability of deep convolutional features learnt by CNN with the excellent classification performance of SVM classifier. Experiments on our product reviews datasets (Amazon Smartphone Review and Taobao Skirt Review), show that the proposed method has higher classification accuracy than plain CNN and SVM.