FREB-TQA

FREB-TQA是一个用于表格问答系统细粒度鲁棒性评估的基准数据集，由博世人工智能中心创建。该数据集包含75,205个实例，主要用于评估表格问答系统在面对表格结构变化、相关单元格内容变化以及数值推理能力时的鲁棒性。数据集通过七种不同的扰动方法生成，并辅以人工标注，以确保评估的准确性和可靠性。FREB-TQA的应用领域主要集中在提高表格问答系统的鲁棒性，解决现有系统在面对复杂和变化环境时的不足。

FREB-TQA is a benchmark dataset for fine-grained robustness evaluation of table question answering (TQA) systems, created by the Bosch Center for Artificial Intelligence. It contains 75,205 instances, and is primarily used to evaluate the robustness of table QA systems when confronted with table structure changes, relevant cell content modifications, and numerical reasoning tasks. The dataset is generated through seven distinct perturbation methods, and is supplemented with manual annotations to ensure the accuracy and reliability of the evaluation. The main application of FREB-TQA is to enhance the robustness of table QA systems and address the shortcomings of existing systems when facing complex and dynamic environments.