Czech Dataset for Cross-lingual Subjectivity Classification

捷克跨语言主观性分类数据集是由西波西米亚大学应用科学学院创建的，包含10000个从电影评论和描述中手动标注的主观和客观句子。该数据集旨在与现有的英语数据集结合，作为基准测试预训练多语言模型在捷克语和英语之间的知识转移能力。数据集由两名标注者进行标注，Cohen’s κ的标注者间一致性达到0.83，是目前首个针对捷克语的主观性数据集。此外，还创建了一个包含200000个自动标注句子的额外数据集。数据集主要应用于解决跨语言主观性分类问题，特别是在零射击跨语言分类任务中评估模型的性能。The Czech cross-lingual subjectivity classification dataset was developed by the Faculty of Applied Sciences, University of West Bohemia. It comprises 10,000 manually annotated subjective and objective sentences extracted from film reviews and descriptions. This dataset is designed to be paired with existing English datasets as a benchmark for evaluating the knowledge transfer capacity of pre-trained multilingual models across Czech and English. The dataset underwent annotation by two annotators, achieving an inter-annotator agreement score of Cohen’s κ = 0.83, marking it the first dedicated Czech-language subjectivity dataset to date. Additionally, a supplementary dataset containing 200,000 automatically annotated sentences was constructed. This dataset is primarily utilized to address cross-lingual subjectivity classification tasks, with a specific focus on evaluating model performance in zero-shot cross-lingual classification tasks.