Image Recommendation Based on Pre-trained Deep Learning and Similarity Matching

Recommender systems are widely used in many domains, especially in E-commerce. It can be used for attracting users by recommending appropriate products to them. There are many techniques in recommendation systems which can predict rating scores to recommend next products. In this work, we propose an approach for recommendation based on product images using pre-trained deep learning models and similarity matching. Specifically, in the proposed model, we have utilized the pre-trained deep learning models (e.g., the VGG16) to extract the image features. Then, based on the image features, we compute similarities between the products (e.g. using Cosine similarity). For recommending similar products to the users, we do the same tasks, e.g., extracting feature of the current image, computing its similarity with other images in the database, and generating a list of TOP-N (e.g. TOP-5) most similar products. Experimental results on two public data sets show that the approach can give good recommendations at more than 90% of accuracy.