Julian McAuley (UC San Diego)

Understanding the Visual Dynamics of Opinions and Purchase

Services such as Amazon, Yelp, Netflix etc. aim to personalize content to users by means of so-called “recommender systems.” Such systems work by modeling users’ previous activities (ratings, purchases, clicks etc.) in order to estimate how they will respond to new content in the future. In this talk, I’ll give a high-level introduction to such systems, and discuss in particular how the visual appearance of items can be modeled for applications such as clothing recommendation. Understanding the visual dimensions of people’s opinions shall help to provide better recommendations, but also to understand how visual preferences (i.e., “fashions”) have changed over time.