Serendipitous Exploration of Large-scale Product Catalogs

Online shopping has developed to a stage where catalogs<br> have become very large and diverse. Thus, it is a challenge<br> to present relevant items to potential customers within a very few<br> interactions. This is even more so when users have no defined<br> shopping objectives but operate in an opportunistic mindset. This<br> problem is often tackled by recommender systems. However,<br> these systems rely on consistent user interaction patterns to<br> predict items of interest. In contrast, we propose to adapt the<br> classical information retrieval (IR) paradigm for the purpose<br> of accessing catalog items in a context of un-predictable user<br> interaction. Accordingly, we present a novel information access<br> strategy based on the notion of interest rather than relevance. We<br> detail the design of a scalable browsing system including learning<br> capabilities joint with a limited-memory model. Our approach<br> enables locating interesting items while not requiring good quality<br> descriptions within a few steps. Our system allows customer to<br> seamlessly change browsing objectives without having to start<br> explicitly a new session. An evaluation of our approach based on<br> both artificial and real-life datasets demonstrates its efficiency in<br> learning and adaptation.