RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation

Zhiyi Duan; Hongyu Yuan; Rui Liu

cs.AIApr 13, 2026

RAG-KT: Cross-platform Explainable Knowledge Tracing with Multi-view Fusion Retrieval Generation

Authors: Zhiyi Duan, Hongyu Yuan, Rui Liu

Organizations: Inner Mongolia University, Hohhot, China

Abstract

Knowledge Tracing (KT) infers a student's knowledge state from past interactions to predict future performance. Conventional Deep Learning (DL)-based KT models are typically tied to platform-specific identifiers and latent representations, making them hard to transfer and interpret. Large Language Model (LLM)-based methods can be either ungrounded under prompting or overly domain-dependent under fine-tuning. In addition, most existing KT methods are developed and evaluated under a same-distribution assumption. In real deployments, educational data often arise from heterogeneous platforms with substantial distribution shift, which often degrades generalization. To this end, we propose RAG-KT, a retrieval-augmented paradigm that frames cross-platform KT as reliable context constrained inference with LLMs. It builds a unified multi-source structured context with cross-source alignment via Question Group abstractions and retrieves complementary rich and reliable context for each prediction, enabling grounded prediction and interpretable diagnosis. Experiments on three public KT benchmarks demonstrate consistent gains in accuracy and robustness, including strong performance under cross-platform conditions.

Figures & tables

Figure 1: Comparison of KT paradigms. Traditional models struggle with cross-platform shifts, while prompt-based LLMs are prone to hallucinations. RAG-KT overcomes these by leveraging multi-view retrieval to enable reliable, grounded inference.

Figure 2: The RAG-KT framework, notation like $S_{tgt}\textendash K_{tgt}$ indicates the target student’s performance on the target concept, while $S_{sim}\textendash K_{tgt}$ represents similar students’ aggregated performance on that dimension. $QG$ is a collection of semantically equivalent $Q$ from different data sources.

Type	Baselines	ASSIST09			ASSIST12			DBE-KT22
Type	Baselines	ACC	AUC	F1	ACC	AUC	F1	ACC	AUC	F1
DL-based Methods	DKT	72.18	81.82	78.12	70.86	69.12	74.98	71.08	72.26	74.52
	DKT+	72.64	82.47	78.65	71.34	70.16	75.42	71.76	73.42	75.06
	AT-DKT	72.62	82.96	79.16	71.58	71.28	75.84	72.32	74.18	75.48
	DKVMN	71.32	81.28	77.92	70.51	68.73	74.76	71.47	72.03	74.65
	Deep-IRT	71.96	80.98	77.84	71.02	69.19	75.03	72.04	72.81	75.09

Table 1: Main results on three benchmark datasets. Bold values denote the best performance; bold and underlined values indicate the second-best.

\text{RAG-KT}_{\text{LLM}}

employs different LLMs as the final prediction model.

Methods	ASSIST09			ASSIST12			DBE-KT22
Methods	ACC	AUC	F1	ACC	AUC	F1	ACC	AUC	F1
Full	80.00	85.74	85.75	74.80	73.89	82.60	78.89	76.32	86.80
w/o Similar Students	77.28	81.07	83.60	69.53	71.47	78.13	77.40	73.98	84.17
w/o Question Groups	76.73	84.25	83.32	70.01	72.33	78.67	77.94	75.50	85.16
w/o KC Interest Subgraph	77.46	85.49	84.23	70.01	73.06	78.54	77.40	75.23	85.08
w/o Structural Path Subgraph	78.50	84.16	84.89	72.40	71.54	80.67	76.71	75.98	85.32

Table 2: Ablation results on three benchmark datasets. Each row removes or alters a key component of the full RAG-KT framework to assess its impact.

Figure 3: Parameter sensitivity analysis of RAG-KT on three datasets.

Type	Baselines	Eedi
Type	Baselines	ACC	AUC	F1
DL-based Methods	DKT	49.54	50.23	49.61
	DKT+	49.98	50.62	49.84
	DKVMN	51.58	51.20	52.64
	AKT	53.55	49.69	47.68
LLM-based Methods (Cold-start)	LOKT	6 5.88	6 5.52	5 7.90

Table 3: Cold-start Results

Figure 4: A case for our RAG-KT. The student answers this question correctly. The red font represents positive analysis and the green font represents negative analysis.

Baselines	Len.	Exp.	Read.	Edu.	Rig.	Total
CIKT	690	4.22	4.68	4.50	3.83	17.23
EFKT	82	2.70	2.69	2.40	2.26	10.05
HISE-KT	350	4.72	4.48	4.35	4.53	18.08
$\text{RAG-KT}_{\text{DeepSeek-R1}}$	297	4.90	4.56	4.88	4.90	19.24

Table 4: Scores for report quality evaluation on four dimensions: Explainability, Readability, Educational Usefulness, and Rigorousness

Methods	Trainging hours (h)	Inference latency (s)	AUC
EFKT	/	5	61.10
HISE-KT	/	19	84.31
CIKT	24	3	82.27
EPLF	4	1	81.27
LLM-KT	10	3	83.55
2T-KT	20	2.5	82.27

Table 5: Cost and Efficiency

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Source	Target	Accuracy
ASSIST09	ASSIST12	96.33
ASSIST09	DBE-KT22	92.67
ASSIST12	DBE-KT22	96.50

Appendix

Table 6: KC Match Accuracy.

Figure 5: Structured Reliable-Context-based Prompt.

Figure 6: Case Study.

HyperParams	Value	ASSIST09			ASSIST12			DBE-KT22
HyperParams	Value	ACC	AUC	F1	ACC	AUC	F1	ACC	AUC	F1
seq_len	5	76.77	82.03	83.62	69.34	68.54	78.52	76.68	74.59	84.58
	10	78.48	8 3.78	85.15	70.37	72.55	78.32	78.19	75.62	85.81
	25	80.00	85.74	85.75	70.37	70.86	79.28	78.89	76.32	86.80
	50	78.28	82.82	85.14	7 4.80	7 3.89	8 2.6	7 7.98	7 6.84	85.59
	100	7 9.90	83.04	8 6.04	75.83	75.75	82.76	78.89	78.72	8 6.60

Appendix

Table 7: Complete HyperParams results.

Dimension	Score	Description
Explainability	1	Lacks any reasoning or presents a chaotic causal chain.
	2	Vague causal logic that is difficult to understand.
	3	Partially valid reasoning but lacks overall coherence.
	4	The reasoning chain is largely complete and causal relationships are clear.
	5	Fully reveals the reasoning process, accurately explaining why a prediction was made.
Readability	1	Chaotic, obscure, or incomprehensible language.

Appendix

Table 8: Scoring Mechanism.

Knowledge Tracing

Knowledge tracing models how learners’ mastery of skills changes over time from their interactions with educational tasks.

See all 31 papers

Explainable Knowledge Tracing via Probabilistic Embeddings and Pattern-based Reasoning
May 10, 2026Siyu Wu, Cong Xu, Wei ZhangRepresentation Learning Interpretable ML
1Shanghai Institute of AI Education, East China Normal University, Shanghai, 200241, China · 2Shool of Computer Sicence and Technology, East China Normal University, Shanghai, 200241, China
MERIT: Memory-Enhanced Retrieval for Interpretable Knowledge Tracing
Mar 3, 2026Runze Li, Kedi Chen, Guwei Feng +3Knowledge Tracing Educational Technology
School of Computer Science and Technology, East China Normal University, China · Shanghai Innovation Institute, China · WeChat AI, Tencent, China
Faster, Cheaper, More Accurate: Specialised Knowledge Tracing Models Outperform LLMs
Mar 3, 2026Prarthana Bhattacharyya, Joshua Mitton, Ralph Abboud +1Knowledge Tracing AI in Education
Eedi · Learning Engineering Virtual Institute

May 10, 2026J/K move · Enter open · S save

Explainable Knowledge Tracing via Probabilistic Embeddings and Pattern-based Reasoning

Siyu Wu, Cong Xu, Wei Zhang

1Shanghai Institute of AI Education, East China Normal University, Shanghai, 200241, China · 2Shool of Computer Sicence and Technology, East China Normal University, Shanghai, 200241, China

Representation Learning Interpretable ML Knowledge Tracing

Open paper AskPDF arXiv

Knowledge Tracing (KT) models students' knowledge states based on learning interactions to predict performance. While deep learning-based KT models have boosted predictive accuracy, most models rely on deterministic vector embeddings and opaque latent state transitions, limiting interpretability regarding how specific past behaviors influence predictions. To address this limitation, we propose Probabilistic Logical Knowledge Tracing (PLKT), an interpretable KT framework that formulates prediction as a goal-conditioned evidence reasoning process over historical learning behaviors. Instead of representing knowledge states as deterministic vector embeddings, PLKT employs robust Beta-distributed probabilistic embeddings to represent student knowledge states. This probabilistic foundation allows us to model the uncertainty of historical behaviors and perform explicit logical operations (e.g., conjunction), constructing transparent reasoning paths that reveal how specific past interactions contribute to the prediction. Extensive experiments show that PLKT outperforms state-of-the-art KT methods while achieving superior interpretability. Our code is available at https://anonymous.4open.science/r/PLKT-D3CE/.

Abstract

Knowledge Tracing

Explore similar work

Explainable Knowledge Tracing via Probabilistic Embeddings and Pattern-based Reasoning

MERIT: Memory-Enhanced Retrieval for Interpretable Knowledge Tracing

Faster, Cheaper, More Accurate: Specialised Knowledge Tracing Models Outperform LLMs