Interpretable Large-Scale Graph Network Model: A Computational Platform for Predicting Instructing Achievement Bias

Abstract

We introduce an interpretable and scalable graph-based system designed to predict biases in instructional achievement across diverse educational data. Aimed at large-scale learning platforms, the system identifies distinctive patterns of instructor-related biases by learning hierarchical features from a wide range of behaviorsąłincluding teaching styles, student interactions, and prior performanceąłusing weakly supervised methods to capture broad influencing factors. To refine these inputs, we apply a geometric feature selection process that removes less informative attributes, improving the quality of instructor and student representations. These refined features are embedded into a latent space where each instructor is characterized by a set of underlying bias-related behaviors. This facilitates a more nuanced understanding of disparities in achievement. Using these representations, we build a weighted similarity graph of instructors, connecting them based on shared bias-related patterns. Advanced clustering algorithms detect meaningful communities within this graph, grouping instructors who exhibit similar instructional biasesąłsuch as preferences for certain teaching methods or styles of student engagement. These communities reveal bias patterns at both broad and detailed levels within teaching environments. To enhance interpretability, the system links these community results with individual instructor profiles, offering personalized insights about instructional biases. By combining local instructional behaviors with broader educational trends, the system improves its ability to predict and address achievement disparities. We tested this approach on datasets containing millions of student and instructor interactions. The results demonstrate that the system scales efficiently and maintains high accuracy across different teaching contexts and instructor activity levels. Its ability to uncover hidden instructional biases and provide actionable feedback makes it a valuable tool for promoting fairness in education.