ECP-Bench:用基础模型评测娱乐内容推广
原标题:ECP-Bench: Benchmarking and Learning Entertainment Content Promotion with Foundation Models
AI 摘要
索尼集团联合KAIST和密歇根大学提出ECP-Bench,用于评测基础模型在娱乐内容推广任务上的能力,并训练了ECP-4B和ECP-9B模型。评测覆盖游戏、电影、音乐等领域的推广技能,包括内容理解、市场预测和用户建模等。结果显示专用微调模型在整体准确率上优于GPT5.5、Gemma4-31B等前沿模型,并在领域泛化和未见任务泛化上表现更佳。
正文节选
[Page 1] ECP-Bench: Benchmarking and Learning Entertainment Content Promotion with Foundation Models Hyomin Kim1,2,*, Bowen Chen1,*, Jin Huang3,*, Zhao Wang1,†, Qiaozhu Mei3, Shingo Takamatsu1 1Sony Group Corporation 2KAIST 3University of Michigan Collected Data for Entertainment Content Promotion Promotion Skill Families Evaluation Highlights :Frontier :Open :Finetuned