Image-based Information Filtering to Compare and Select Items

Current recommender systems overlook the role of images in conveying information about items, focusing on metadata, ratings, and reviews for the generation and presentation of the suggestion lists. However, images describe different aspects of an item, which might be used to steer its presentation to satisfy specific information needs. We propose two user interfaces for image-based information filtering that support item exploration and comparison by analyzing the scenes described by the images, or the objects recognized in them. In a user test in the home-booking domain, we found that participants preferred scene-based information filtering to object-based and traditional visualization of items and reviews in product catalogs.