A Deep Learning Based Hybrid Model for Sales Prediction of E-commerce with Sentiment Analysis
Haotian Zhu
Abstract
The market of E-commerce has developed rapidly since the emergence of Internet. Many companies or shops of E-commerce want to seek a way to predict the sales of their products. The prediction of sales can help the merchants to formulate a sales strategy, in order to obtain a bigger profit and attract more investment. However, many studies simply use the daily sales or some very basic daily information of the product to predict the sales, without considering the effects of reviews. In this paper, a hybrid network including Bi-directional Long Short-Term Memory (BiLSTM) and Convolutional Neural Network (CNN) is proposed to solve this prediction task. Through careful selection, several attributes and comments of the products are utilized. Feature engineering is used to normalize the different kinds of data. BiLSTM is conducted to analyze all the comments. CNN is utilized to make predictions by using the data provided by feature engineering. Analysis is described to show the advantages and effectiveness of this network. By using the attributes of the product, along with the polarization of comments, CNN can predict the sales of the product.