AAKT: Enhancing Knowledge Tracing With Alternate Autoregressive Modeling

2025-02-17

IEEE Transactions on Learning Technologies

AI Summary

To address insufficient modeling of sequential knowledge-state transitions in knowledge tracing (KT), this paper proposes QAGenKT, an autoregressive generative framework based on alternating question–answer sequences. It formalizes student exercise histories as latent-state evolution governed by interleaved question and answer tokens, jointly encoding prior and posterior knowledge states within a unified latent space. Methodologically, QAGenKT innovatively integrates multi-source educational signals—including skill labels and response times—as auxiliary tasks and implements end-to-end multi-task learning via a Transformer decoder. Evaluated on four real-world KT benchmarks, QAGenKT consistently outperforms state-of-the-art baselines across AUC, accuracy, and RMSE. Ablation studies and visualization analyses validate the critical contributions of (1) question–answer alternating modeling, (2) unified dual-state representation, and (3) multi-source signal fusion.

Technology Category

• Knowledge Representation and Reasoning: Knowledge Acquisition
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Question Answering

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Knowledge tracing (KT) aims to predict students' future performances based on their former exercises and additional information in educational settings. KT has received significant attention since it facilitates personalized experiences in educational situations. Simultaneously, the autoregressive (AR) modeling on the sequence of former exercises has been proven effective for this task. One of the primary challenges in AR modeling for KT is effectively representing the anterior (preresponse) and posterior (postresponse) states of learners across exercises. Existing methods often employ complex model architectures to update learner states using question and response records. In this study, we propose a novel perspective on KT task by treating it as a generative process, consistent with the principles of AR models. We demonstrate that knowledge states can be directly represented through AR encodings on a question–response alternate sequence, where model generate the most probable representation in hidden state space by analyzing history interactions. This approach underpins our framework, termed alternate autoregressive KT (AAKT). In addition, we incorporate supplementary educational information, such as question-related skills, into our framework through an auxiliary task, and include extra exercise details, such as response time, as additional inputs. Our proposed framework is implemented using advanced AR technologies from Natural Language Generation for both training and prediction. Empirical evaluations on four real-world KT datasets indicate that AAKT consistently outperforms all baseline models in terms of area under the receiver operating characteristic curve, accuracy, and root mean square error. Furthermore, extensive ablation studies and visualized analysis validate the effectiveness of key components in AAKT.

Problem

• Enhances knowledge tracing with autoregressive modeling.
• Predicts student performance using exercise sequences.
• Incorporates educational details for improved accuracy.

Innovation

• Alternate Autoregressive Knowledge Tracing
• Generative process for knowledge states
• Incorporates supplementary educational information