Sales Prediction Based on Product Titles and Images with Deep Learning Approaches
Over the past few years, online shopping gradually became the mainstream shopping method. More and more local retailers chose to start their businesses on e-commerce platforms. However, few can survive due to the competitive pressure from the big companies and the entry barriers. The motivation of this project is to identify product listing strategies, primarily visual and textual presentation, that can help retailers to raise their product sales. To achieve that, we build a neural network architecture to predict product sales from text, image and other product listing features that retailers can control. In particular, we use pre-trained VGG16 model for image data and TF-IDF model for text data in our architecture. The accuracy of the model is 73.91%, far above the accuracy of human raters, which is 55.33%.
