Image Based Recommender System using Transfer Learning
Nikhil Kumar Singh, Abhimanyu Kumar · 2022 2nd International Conference on Emerging Frontiers in Electrical and Electronic Technologies (ICEFEET) · 2022
Recommender systems have become extremely com-mon nowadays. These are used in various fields such as e-commerce, video streaming platforms, social media etc. Based on a person's previous history and taste, a person is recommended items. These recommender systems are software tools and techniques that are most commonly used by the companies to target customers. Traditional recommender systems use user feedback such as ratings, likes/dislikes etc to provide recommendations through various techniques such as Content Based Filtering, Collaborative Filtering etc. However there are many instances when a user has interacted with item but has not rated the same item. Also in many cases the rating data is mostly sparse i.e. either the rating is 0 or the rating has not been provided by the user, so the images of items that the users interacted with can be used to provide recommendations. Many e-commerce sites use this method for recommending items to users. In this paper we will be discussing an image recommendation system that can be used to recommend items only using images. It is implemented using transfer learning and cosine similarity, the system attempts to find a result nearest to the image interacted.