SportsOpi: A Novel Dataset for Analyzing Public Sentiment on Controversial Sports Events in YouTube Comments

Sports engages billions of followers worldwide and impacts theeconomy. Sports controversies often ignite passionate discus-sions among fans, analysts, and players. With the rise of socialmedia, platforms like YouTube have become central to these discus-sions. This study aims to analyze the stances or perform opinionmining namely for, against, and neutral on comments from fa-mous social media platforms like YouTube for famous public sportscontroversies. To our knowledge, it is the first-ever study and dataset (hand curated) of civicengagement in controversial sports events spanning around 40 years.LLMs (Llama and Deepseek reasoning family) were used for initialannotations (stance) of comments and later fine-tuned for comparative performance analysis ( 30% boost in accuracy). This dataset presents a collection of YouTube comments (around 43k) on famousand controversial Public Sports Events. We explore public sentiment analysis (stance detection) on a total of 6 famous controversial sports incidents by extracting and processing YouTube comments.Stance detection is performed on those events through fine-tuningof models like Llama-3.1-8b and Deepseek reasoning models (Llama-8b distilled) on comments from events like The Underarm Incident,Jonny Bairstow’s Run-Out Incident, Ashwin’s Mankading Event,Luis Suarez Handball Event etc.体育在全球拥有数十亿受众，并对经济产生显著影响。体育争议事件往往会引发球迷、赛事分析师与运动员群体的热烈讨论。随着社交媒体的兴起，YouTube等平台已成为这类讨论的核心阵地。本研究旨在针对全球知名公共体育争议事件，对YouTube等主流社交媒体平台上的相关评论展开立场分析，即开展观点挖掘（Opinion Mining）任务，识别评论的支持、反对与中立三类立场。 据我们所知，本研究及所构建的人工精选数据集，是全球首个针对跨度近40年的体育争议事件展开公众参与分析的成果。研究首先采用大语言模型（Large Language Model，LLM，涵盖Llama与Deepseek推理系列模型）对评论进行初始立场标注，随后对模型进行微调以开展对比性能分析，最终将模型准确率提升了30%。 本数据集共收录约4.3万条YouTube评论，均来自全球知名的公共体育争议事件。 我们通过提取并处理YouTube评论，对共计6起知名体育争议事件展开公众情感分析，即立场检测（Stance Detection）。研究针对板球腋下投球事件、乔尼·贝尔斯托跑杀争议事件、阿什温曼卡德犯规事件、路易斯·苏亚雷斯手球事件等赛事的评论，对Llama-3.1-8b、Deepseek推理模型（经Llama-8b蒸馏优化）等模型进行微调，以此完成上述事件的立场检测任务。