SS-Bench
收藏资源简介:
SS-Bench是由北京交通大学和腾讯MLPD联合开发的社会故事生成与评估基准数据集。该数据集包含5085个故事,旨在通过严格的结构清晰性、描述性导向和情境安全性约束,帮助自闭症儿童理解和参与日常社交活动。数据集的创建过程采用了一种名为STARSOW的约束驱动策略,通过高级语言模型如GPT-4o生成故事。SS-Bench的应用领域主要集中在自闭症儿童的社交技能干预,旨在通过自动化和可访问的方法提高故事的多样性和时效性,从而改善这一特殊群体的未来生活质量。
SS-Bench is a benchmark dataset for social story generation and evaluation co-developed by Beijing Jiaotong University and Tencent MLPD. This dataset contains 5085 stories, which is designed to help autistic children understand and engage in daily social activities by imposing strict constraints on structural clarity, descriptive orientation and situational safety. The dataset was developed using a constraint-driven strategy named STARSOW, with stories generated by advanced large language models such as GPT-4o. Its main application fields focus on social skill intervention for autistic children, aiming to improve the diversity and timeliness of stories through automated and accessible methods, thereby enhancing the future quality of life for this special group.
数据集概述
数据集名称
SS-Bench: A Benchmark for Social Story Generation and Evaluation
相关论文
- 论文标题:SS-Bench: A Benchmark for Social Story Generation and Evaluation
- 论文链接:https://arxiv.org/abs/2406.15695




