统一跨数据集基准

统一跨数据集基准是由腾讯云与智慧产业事业群和哈尔滨工业大学联合创建，包含14个自然语言推理（NLI）数据集，旨在解决现有NLI数据集存在的标注偏差问题。这些数据集涵盖多种创建协议，如人工引发、人工评判和自动重构，确保数据多样性和广泛性。创建过程中，研究团队采用了跨数据集评估方法，以减少特定数据集偏差对模型评估的影响。该数据集主要应用于自然语言处理领域，特别是NLI模型的泛化性能评估，以推动更可靠的NLI研究发展。

The Unified Cross-Dataset Benchmark was co-developed by Tencent Cloud and Smart Industries Group and Harbin Institute of Technology. It contains 14 natural language inference (NLI) datasets, aiming to address the annotation bias issue prevalent in existing NLI datasets. These datasets cover diverse creation protocols, including human elicitation, human judgment, and automatic reconstruction, to ensure data diversity and broad coverage. During its development, the research team adopted a cross-dataset evaluation method to reduce the impact of dataset-specific biases on model assessment. This benchmark is primarily utilized in the field of natural language processing, particularly for evaluating the generalization performance of NLI models, so as to advance more reliable NLI research.