Jacob Munson (Individual Interdisciplinary PhD Program, MSU) 

11/19/21  2:10pm

Abstract:

Recommender systems are a pervasive technology in the electronic world with applications in e-commerce, online streaming services, social media, news, and more. In this talk, we discuss collaborative filtering, a major paradigm of recommender systems focused on exploiting communal information in large user bases. Within collaborative filtering, we cover types of interactions between recommender systems and users, such as clicks, views, listens, and ratings. Each type of interaction focuses on factoring a large "ratings matrix" with upwards of 90% missing values. We discuss a handful of successful matrix factorization approaches for low rank matrix completion, including bias, probabilistic, and temporal frameworks, with estimation via techniques such as gradient descent and Gibbs Sampling. Special attention is given to the Netflix Prize, a performance driven research effort sponsored by Netflix, which helped to propel the field into the mainstream.