Tag
The paper evaluates LLMs as graders for computer-science exams, finding that prompt design significantly affects grading accuracy, and shows that LoRA fine-tuning can mitigate issues to match human grader performance.
The paper introduces THPT-Ladder, a benchmark that evaluates AI models using Vietnam's 2025 convex marking scheme, revealing how partial credit gaps lead to inaccurate assessments compared to standard accuracy metrics.
This paper describes using a fine-tuned multimodal LLM based on Qwen3.5 to simulate student response probabilities and estimate item difficulty parameters for multiple-choice assessments, approximating 3PL and MCM curves.
This paper proposes a generative AI-assisted summarization framework using GPT-5 model variants to address input-length limitations in automated essay scoring, demonstrating trade-offs between model capacity, summary fidelity, and computational cost on the ASAP 2.0 dataset.
Introduces Epi2Diff, a framework that maps LLM reasoning traces into cognitive episodes to predict human item difficulty, outperforming baselines and providing interpretable process evidence.
Introduces a dataset of 32,534 double-marked GCSE student responses across five subjects, finding that top-performing LLMs agree with examiners more closely than examiners agree with each other, including on handwritten and subjective tasks.
This paper evaluates 42 large language models on their ability to measure item discrimination in reading comprehension assessments, finding weak alignment with human-calibrated measures and highlighting it as an open challenge for psychometric evaluation.