300 High-Myopia Multiple-Choice Questions Generated by Five Large Language Models

300 multiple-choice questions on high myopia generated by five large language models for a benchmarking study. The dataset was created by Ligang Jiang and last updated on May 5, 2026. It includes objective metrics and expert ratings for evaluating the models' performance in generating educational content.

Use Cases

• Benchmarking LLM performance on specialized medical question generation based on the described evaluation blueprint.
• Analyzing the relationship between output features like explanation length and distractor quality based on the reported statistical associations.
• Training or fine-tuning models for educational content creation based on the structured MCQ format and expert ratings.

Strengths

• All 300 generated questions achieved 100.0% initial structural acceptability and format compliance.
• Keyed-answer accuracy was 100.0% for two of the five models tested.
• Expert inter-rater agreement was good, with an ICC range of 0.835–0.885.

Limitations

• The dataset is very small at 27.9 KB, limiting its scope for large-scale analysis.
• Column-level documentation is absent; field semantics must be inferred after download.
• The data is focused on a single, specialized medical topic (high myopia) under standardized Chinese prompting conditions.

Provenance

Source

figshare, authored by Ligang Jiang.

Collection Method

Generated by five LLMs (ChatGPT-5.4, Gemini 3, DeepSeek, Kimi K2.5, Doubao) completing 60 predefined tasks each.

Freshness

Last updated 2026-05-05 05:28:45; freshness should be verified.