MultiSocial
收藏资源简介:
MultiSocial是由肯彭兰切智能技术研究所创建的多语言、多平台数据集,旨在解决社交媒体文本中机器生成文本的检测问题。该数据集包含22种语言,涵盖5个社交媒体平台,总计472,097条文本,其中约58,000条为人类编写,其余由7种多语言大型语言模型生成。数据集的创建过程涉及文本的选择、预处理、生成和后处理,详细信息在附录B中提供。MultiSocial的应用领域主要集中在社交媒体文本的机器生成检测,特别是在零样本和微调形式的检测方法比较中显示出重要价值。
MultiSocial is a multilingual, multi-platform dataset developed by the Kepenglanche Intelligent Technology Research Institute, designed to address the challenge of machine-generated text detection in social media texts. This dataset encompasses 22 languages across 5 social media platforms, with a total of 472,097 text samples. Approximately 58,000 of these samples are human-written, while the remainder are generated by 7 multilingual large language models (LLMs). The creation process of this dataset involves text selection, preprocessing, generation, and post-processing, with detailed information provided in Appendix B. The main application domain of MultiSocial is machine-generated text detection for social media, and it exhibits significant value, especially in comparative studies of detection methods under zero-shot and fine-tuning scenarios.

- 1MultiSocial: Multilingual Benchmark of Machine-Generated Text Detection of Social-Media Texts肯彭兰切智能技术研究所 · 2024年



