Studying Distributed Representations of Pedagogical Objects from Learner Process Data


The aggregate behaviors of students in a learning environment can collectively express nuanced information about the pedagogical objects with which they interact. As these interactions increasingly involve digital components, the availability of data traces of these sequences of interactions at scale increases; and with it, opportunities to study evidence of distributed cognition. In this talk, we’ll present results surfacing properties of objects from sequences in two different educational contexts; item answering sequences in an online STEM tutoring platform and course selection sequences at Berkeley. A connectionist (neural network) model is used to represent objects as a function of the frequency distribution of other objects also occurring in the same sequence segments, embedding the objects into a vector space, which allows for analogical reasoning about objects in the space. The semantics of the space are explored in the tutoring context by mapping skill meta data of the items to the space, allowing for the skill of items with no meta data to be inferred with high accuracy. In the University context, the semantics of the space is populated using course description text, revealing fine-grained topics of courses not present in their course descriptions but observable through these analyses due to the distributed representation of those topics in the vector space. Other affordances of this paradigm of representation will be discussed, including its applicability to enhanced course search and course articulation.

Zachary Pardos is an assistant professor in the Graduate School of Education at UC Berkeley with a joint appointment in the School of Information. His research focuses on knowledge representation and personalized supports leveraging big data from educational contexts. He directs the Computational Approaches to Human Learning research lab and is the recent recipient of an NSF AI educator award.

Tuesday, February 27, 2018 - 3:30pm