FlatFormer: A Flat Transformer Knowledge Tracing Model Based on Cognitive Bias Injection

TL;DR

FlatFormer introduces a simplified flat Transformer architecture with cognitive bias injection mechanisms, achieving state-of-the-art knowledge tracing performance with reduced complexity and faster inference.

Contribution

The paper presents a novel flat Transformer model with information injection techniques, avoiding hierarchical complexity while maintaining high cognitive fidelity in knowledge tracing.

Findings

Achieves state-of-the-art AUC on large-scale datasets.

Uses less than 15% of parameters compared to hierarchical models.

Increases inference speed by approximately three times.

Abstract

Knowledge Tracing (KT) models face a critical ``Performance-Complexity Trap'': capturing complex cognitive dynamics like learning sessions and memory decay typically requires deep hierarchical architectures, which incur prohibitive computational costs for real-time deployment. To resolve this, we propose FlatFormer, a streamlined architecture based on the novel design paradigm of ``Information Injection over Structural Stacking.'' Unlike parameter-heavy hierarchical models, FlatFormer leverages a standard flat Transformer augmented with two lightweight injection mechanisms: (i) a hybrid input encoding strategy combining learnable session identifiers with fixed sinusoidal step embeddings; and (ii) a pre-computed power-law bias integrated directly into attention logits to explicitly model the forgetting curve. Extensive experiments on four large-scale datasets (e.g., EdNet, Junyi) show that FlatFormer achieves state-of-the-art performance. For example, on the EdNet dataset, compared to the strongest hierarchical baseline (HiTSKT), its absolute AUC increased by 8.3%, while using less than 15% of parameters, and inference speed was about three times faster. These results validate that high cognitive fidelity does not necessitate architectural complexity.