A Dual-Annotated YouTube Comment Corpus on Precision Fermentation Technology (2019–2024)

This dataset contains 16,652 YouTube comments related to precision fermentation technology, collected via the YouTube Data API from 180 videos spanning February 2019 to November 2024. Each comment has been independently classified for sentiment (positive, negative, or neutral) by two models: GPT-4o (OpenAI) and EmoRoBERTa (a RoBERTa-based transformer fine-tuned on GoEmotions). The two models disagree on 39.1% of classifications, with EmoRoBERTa systematically underdetecting negative sentiment relative to GPT-4o. The deposit contains two files: human_annotated_subset.csv (n = 301): A subset with sentiment labels from both models plus a human annotator. Fields: VideoID, CommentID, Timestamp, GPT4o_Label, EmoRoBERTa_Label, Human_Label. full_corpus.csv (n = 16,652): The complete corpus with dual model annotations. Fields: VideoID, CommentID, Timestamp, Reply_Flag, GPT4o_Label, EmoRoBERTa_Label. Comment text is excluded to comply with YouTube API Developer Policies. Researchers may retrieve original comment text using the provided CommentIDs via the YouTube Data API. The data collection, preprocessing, and classification code is available at https://github.com/isom-ds/ffsp-sm-emotions.本数据集包含16652条与精准发酵技术相关的YouTube评论，通过YouTube Data API从2019年2月至2024年11月期间的180个视频中采集得到。每条评论已由两个模型独立完成情感分类（分类类别为积极、消极或中性）：GPT-4o（OpenAI）以及EmoRoBERTa——一种基于RoBERTa的Transformer模型，在GoEmotions数据集上微调得到。两个模型在39.1%的分类结果上存在分歧，且相较于GPT-4o，EmoRoBERTa系统性地漏检消极情感。 本数据集包含两个文件： 1. human_annotated_subset.csv（样本量n=301）：包含两个模型的情感标签以及人工标注者标注结果的子集，字段包括：VideoID、CommentID、Timestamp、GPT4o_Label、EmoRoBERTa_Label、Human_Label。 2. full_corpus.csv（样本量n=16652）：包含双模型标注结果的完整语料库，字段包括：VideoID、CommentID、Timestamp、Reply_Flag、GPT4o_Label、EmoRoBERTa_Label。 为遵守YouTube API开发者政策，本数据集未包含评论原文。研究人员可通过给定的CommentID调用YouTube Data API获取原始评论文本。数据采集、预处理及分类代码已开源至https://github.com/isom-ds/ffsp-sm-emotions。