Recommender systems (RS) have emerged as a popular means for helping users discover items of interest such as news stories and posts, as well as a platform with which to push products that will likely appeal to users, hopefully leading to a purchase. Most of the emphasis on traditional RS research has been placed on prediction accuracy, and other measures such as diversity and serendipity.
Recently, there has been a growing body of work on optimizing metrics normally considered external to a RS. Examples include recommending packages, item exchanges, and profit maximization to name a few. In this talk, I will present two such optimization problems. The first one asks, can we exploit an existing operational RS to launch a marketing campaign for a product? Specifically, can we select a set of “seed” users for a marketing campaign for a new product, such that if they endorse the product by providing relatively high ratings, the number of other users to whom the product is recommended by the underlying RS is maximum.
The second problem is related to group recommendations. The main focus in group recommendations has been on generating recommendations for given groups that maximize the group satisfaction under standard semantics like least misery or aggregate voting. Instead of relying on ad hoc group formation, we ask, can we form groups strategically, such that group recommendations produced by existing algorithms will lead to maximum group satisfaction?
I will discuss the motivating applications for both problems, their hardness, and approximation, as well as the results of our experiments on several real datasets. I will conclude with directions for future research.