Learner models fall into two categories: (1) Domain Models, which associate skills with tasks to uncover the structure of the skill domain, and (2) Knowledge Models (also called Knowledge Tracing models), which track how a student's knowledge changes while practicing. For a Knowledge Tracing (KT) model to perform well, it must accurately represent the student's knowledge state and predict whether the student will answer a given problem correctly. Existing KT models struggle to represent knowledge state or knowledge interaction — the interaction between a student's knowledge state at time $t$ and a skill at time $t+1$. To address this gap, the paper introduces the Knowledge Query Network (KQN).