Tag-based Personalized Image Ranking In Event Browsing

Many image sharing websites, e.g. Flickr, Google+, allow users to upload imagesas an event, and users can browse the images others uploaded as events. The fact thatpeople usually browse only the frst few images of an event then decide whether theevent is what they want makes us believe that it is necessary to present those imagespeople favor on the very frst position for each event. Here we propose a new tag-basedpersonalizedimage-rankingalgorithmineventbrowsingsuchthatitgivesimagehigherscore if it: a) is important in the event, b) matches user’s preference. c) matches user’s query. To this end, we frst adopt a local matching model to assign images an originalscore based on whether this image satisfes user’s query and preference. We thenpropose a global ranking model to take the local scores as initial values and make thesalience scores iteratively smooth with respect to all images returned from the eventsof the query.