Online shopping has developed in parallel with the Internet, and Recommendation Systems have played a pivotal role in its growth. The recommendations are usually provided in two ways: Content-based Filtering and Collaborative Filtering. Both forms of recommendations face the problem of Cold-Start due to an initial lack of information. To overcome this issue, Image-based Recommendation Systems are introduced in order to allow the users to locate products based on similarity of images when purchasing products in categories such as: clothes, shoes, home-decor, kitchen and dining utilities, jewelry, and accessories by mostly viewing images. In this thesis, a Hybrid Model of displaying similar images to that of the product being viewed was developed using Deep Features and Description-based Models. The Hybrid Model displayed a set composed of all images that belong to both Deep Features and Description-based Models. Implementation and comparison of results were performed on 100,000 images of SBU Captioned Photo Dataset.
展开▼
