PhD Dissertation Defense by DO Ha Loc | Mining Diverse Consumer Preferences For Bundling And Recommendation

Mining Diverse Consumer Preferences For Bundling And Recommendation

Consumers have diverse preferences. That they share similar tastes on some products does not guarantee their agreement on other products. In this talk I present my study on this diversity from two perspectives, namely 1) between consumers and 2) between products.

Diversity of preferences between consumers is studied in the context of recommendation systems. Previous approaches assume a singular value to capture how two consumers are similar in their tastes. We relax this assumption by proposing a method measuring different degrees of similarity between consumers. Specifically, we propose a probabilistic framework CAM-DPMF to capture how likely two consumers share similar preferences on particular products based on their observed rating data.

Diversity of preferences between products is studied in the context of product bundling. Bundling is a marketing strategy to offer a combination of products (i.e., bundle) at one price. Since it is impractical to offer all possible bundles to market, firms are interested in a selective set of bundles satisfying some objectives and constraints. Finding such optimal bundle set is computationally challenge due to the large search space of bundle set. On this second half of the talk, I would briefly discuss computationally efficient approaches to search for profit-maximizing bundle sets in two forms, namely bundling configuration and top-K bundle.