Enhancing Directed Acyclic Graphs for Reliable Causal Discovery in Education

Abstract

Discovering causal relationships in educational data is critical, yet traditional directed acyclic graph (DAG) construction methods often fail with noisy data. We propose the Causal Annotation Platform (CAP), which integrates teacher expertise into algorithmic causal discovery. CAP visualizes DAGs, simulates interventions via do- calculus, and incorporates expert annotations as weighted constraints in structure learning. Our prototype demonstrates how qualitative teacher judgments can be transformed into quantitative algorithmic inputs. By using expert input as algorithmic constraints, the system addresses limitations of conventional data-driven approaches. Future studies will evaluate the effectiveness of this method.