From Diagnosis to Dialogue: A Knowledge-Augmented Framework for Generating Interpretable Feedback from Deep Knowledge Tracing

• In Event: Virtual Poster Session
  In Virtual iPresentation Only: SIG-Measurement and Assessment in Higher Education Virtual Poster Session

Abstract

The precise assessment of AI talent is critical for personalized cultivation. While Deep Knowledge Tracing (DKT) effectively models knowledge states, its probabilistic outputs lack the interpretability for actionable feedback. To address this, our study proposes an innovative framework designed specifically for AI talent assessment. It employs a structured assessment based on Bloom’s Taxonomy and a DKT model for dynamic diagnosis. The core innovation is a Knowledge-Augmented Generation (KAG) module, where a Large Language Model (LLM) contextualizes DKT’s outputs using an AI discipline Knowledge Graph that maps conceptual dependencies. This synthesis enables reports that elucidate root causes of learning gaps and provide personalized pathways, allowing for more targeted guidance for developing AI talent.

Authors

• Zhen Li, Central China Normal University

• Chenxuan He, Central China Normal University