SRJudge consistently outperforms state-of-the-art baselines on three knowledge concept tagging datasets¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? SRJudge achieves the best F1 scores of 0.7602 on S Math, 0.6987 on S Bio, and 0.6643 on S Phy, outperforming the leading baseline LGCEL. → Zhiwei Yang 2026

Evidence¶

Zhiwei Yang 2026¶

Zhiwei Yang, Jiahua Yang, Huiru Lin, Xing Chen, and Quanlong Guan. (2026). SRJudge: Empowering Large Language Models with Selective Reasoning for Fine-Grained Knowledge Concept Tagging. arXiv preprint. https://arxiv.org/abs/2609.36982 doi:10.48550/arxiv.2609.36982

q2 · i? · design · r2

Benchmark comparison on three educational datasets (S Math, S Bio, S Phy) with 8:1:1 stratified splits, evaluating accuracy, precision, recall and macro F1 against SLM, LLM and task-specific baselines. The article reports SRJudge "consistently yielding the best results", with LGCEL the leading baseline at F1 0.7311, 0.6840 and 0.6476 respectively.

"Overall, SRJudge outperforms LGCEL, achieving an F1 score of 0.7602 on S Math, 0.6987 on S Bio, and 0.6643 on S Phy, consistently yielding the best results."

Discussion¶

Related Claims¶

• Applying GRPO alone to LLMs for tagging yields significantly lower performance than supervised fine-tuning — related
• Each SRJudge component contributes: the Reasoner and Judger add measurable F1 gains over the Selector alone — related
• SRJudge improves prediction on hard-to-distinguish samples, with larger gains on S Math than on the higher-grade datasets — related
• A dynamic position reward in RL training improves tagging performance and prevents overfitting to Selector outputs — related
• Pruning low-advantage completions during RL improves F1 up to a middle pruning rate while reducing training time — related