plosone_minimal_dataset.csv

This dataset contains the results of a novel, LLM-driven annotation process applied to expert coffee reviews, as presented in the manuscript "Automated Multi-Label Coffee Flavor Classification: A Comparative Study of BERT and TF-IDF using LLM-Driven Data Annotation."This "Minimal Dataset" is provided to ensure the reproducibility of our findings. It includes:A unique <code>review_id</code> for each entry.The original <code>blind_assessment</code> text used as input for the LLM.The original quantitative sensory scores (<code>Final Score</code>, <code>Aroma</code>, <code>Acidity/Structure</code>, etc.) provided by human experts, which were used for the quantitative validation of the LLM's annotations.The final 17 columns of binary (0/1) flavor labels as generated by the LLM.

本数据集收录了一项将新型大语言模型（LLM）驱动的标注流程应用于专业咖啡品鉴评论的实验结果，相关研究成果已发表于论文《自动多标签咖啡风味分类：基于大语言模型驱动数据标注的BERT与TF-IDF对比研究（Automated Multi-Label Coffee Flavor Classification: A Comparative Study of BERT and TF-IDF using LLM-Driven Data Annotation）》。本「极简数据集（Minimal Dataset）」旨在保障研究结论的可复现性，其包含以下内容：1. 每条数据独有的<code>review_id</code>（评论ID）；2. 用于输入大语言模型的原始盲评文本<code>blind_assessment</code>；3. 人类专家提供的原始感官定量评分，包括<code>Final Score</code>（最终得分）、<code>Aroma</code>（香气）、<code>Acidity/Structure</code>（酸度/风味结构）等，该评分用于对大语言模型生成的标注结果进行定量验证；4. 大语言模型生成的共17列二元（0/1）风味标签。