The classic “Amazon style” recommendation based on “users like you” has been successful in e- commerce. In educational contexts, it is insufficient for supporting students in achieving their goals and introducing them to the new and unexpected. In this talk, I will discuss work published at ACM RecSys and Learning Analytics and Knowledge (LAK) describing the unique student and institutional values that necessitate special consideration and modifications to the common recommender systems approach. A production course guidance system deployed at UC Berkeley (<http://askoski.berkeley.edu>) and within-course recommender systems extending edX functionality will be demonstrated as well as the neural network-based algorithmic modifications made to implement them at scale and adapt them to the domain.
Dr. Zachary Pardos, Assistant Professor at UC Berkeley, is a leading learning scientist and an internationally recognized expert in educational data mining and learning analytics. He earned his PhD in Computer Science at Worcester Polytechnic Institute, where he worked with K-12 educators and students to integrate educational technology into the curriculum as a formative assessment tool. He holds several academic leadership positions in the learning analytics community, including posts as an editorial board member for two of its journals and executive committee member for the AI in Education Society. He has published over 50 peer-reviewed articles on topics covering student knowledge representation and recommender systems in higher ed. As a faculty member in the Graduate School of Education and Information at Cal, he teaches courses on data mining, machine learning, and digital learning platforms and directs the computational approaches to human learning lab. Prior to joining Berkeley, he was a postdoctoral associate at MIT in the Computer Science AI Lab and Research Lab for Electronics. Twitter: @zpardos