iis-research-team/AINL-Eval-2025

AINL-Eval 2025共享任务数据集旨在检测俄语科学摘要中的人工智能生成内容。该数据集包含了使用不同语言模型生成的文本，分为训练集、公开测试集和私有测试集。训练集包含来自10个领域的约35,000个文本，每个文本都标记为来自人类或特定模型。公开测试集包括相同领域的文本以及一个未见过的模型的文本。私有测试集包含来自10个不同领域的约6,000个文本，其中8个领域与训练集重叠，另外2个领域是新领域，文本来源包括人类编写的和上述模型以及一个未知模型生成的文本。

The AINL-Eval 2025 Shared Task dataset is aimed at detecting AI-generated scientific abstracts in the Russian language. The dataset consists of texts generated by various language models and is divided into a training set, a public test set, and a private test set. The training set includes approximately 35,000 texts from 10 domains, each labeled as originating from a human or a specific model. The public test set includes texts from the same domains and an additional unseen model. The private test set contains approximately 6,000 texts from 10 different domains, with 8 overlapping with the training set and 2 new domains, including texts written by humans and generated by the aforementioned models and an unknown model.