A Product Recommendation Method by Analyzing Images and Reviews in E-Commerce

This paper proposes a method for the product visual search service in e-commerce to recommend products with high satisfaction by computing image scores and review scores. To compute image scores, we calculate the image similarities between user-inputted images and product representative images, and the text similarities between image labels and product descriptions. To compute review scores, we perform a sentiment analysis on product reviews and combine with user repeated purchase behavior to recommend products with high satisfaction. Finally, we evaluated and discussed our proposed method based on a visual search using real product data.