Recent advances in large reasoning models (LRMs) have demonstrated impressive capabilities in highly structured domains such as mathematics and programming. However, their application to education-where effective reasoning must be pedagogically meaningful, context-sensitive, and responsive to real student needs-remains relatively unexplored. Existing large language models (LLMs) often struggle to deliver instructional coherence, formative feedback, or simulate sophisticated teacher decision-making, limiting their practical utility in educational settings. To fill this gap, we present Pedagogy-R1, a comprehensive pedagogical reasoning framework designed to adapt LLMs for authentic classroom tasks. Our approach features three key innovations: (1) a distillation-based training pipeline that uses pedagogically filtered outputs for instruction tuning, (2) the Well-balanced Educational Benchmark (WBEB), which systematically evaluates models across five dimensions-subject knowledge, pedagogical knowledge, knowledge tracing, essay scoring, and real-world teacher decision-making-and (3) the Chain-of-Pedagogy (CoP) prompting strategy, employed both to generate pedagogically enriched training data and to elicit teacher-like reasoning during inference. We conduct a mixed-methods evaluation, combining fine-grained quantitative analyses of model performance with qualitative insights into the model's pedagogical reasoning patterns.

This paper presents novel techniques for enhancing the performance of knowledge tracing (KT) models by focusing on the crucial factor of question and concept difficulty level. Despite the acknowledged significance of difficulty, previous KT research has yet to exploit its potential for model optimization and has struggled to predict difficulty from unseen data. To address these problems, we propose a difficulty-centered contrastive learning method for KT models and a Large Language Model (LLM)-based framework for difficulty prediction. These innovative methods seek to improve the performance of KT models and provide accurate difficulty estimates for unseen data. Our ablation study demonstrates the efficacy of these techniques by demonstrating enhanced KT model performance. Nonetheless, the complex relationship between language and difficulty merits further investigation.