Scalable image-based search-and-discovery

People use online search-and-discovery services, such as Yelp, by first finding a specific item with keywords and then examining the images linked to the item. Images could constitute an important part of users' decision-making process but users reach them indirectly. Although recently researchers proposed several image-based search interfaces, how they can effectively arrange huge number of images in a scalable manner is still not clear. To address this, we introduce PicNav, an image-driven navigation system that automatically arranges photos according to their semantic similarity. PicNav is built on deep neural networks learned from the Yelp food dataset and enables effective zoomin/out features. We conducted interviews with ten users to qualitatively assess the system's usability. The users identified a number of advantages of PicNav, providing insights into the general use of imagery in search-and-discovery services.