OpenPI-C

OpenPI-C数据集是由伊利诺伊大学厄巴纳-香槟分校创建，专注于开放词汇状态跟踪任务。该数据集通过筛选和清洁原始OpenPI数据集中的低质量数据构建而成，包含13,800条状态变化记录。数据来源于WikiHow，经过三阶段的人工清洁过程，确保数据质量。OpenPI-C旨在解决模型在处理开放词汇状态跟踪时的准确性和可靠性问题，特别是在避免模型产生幻觉方面具有重要应用。The OpenPI-C dataset, developed by the University of Illinois Urbana-Champaign, focuses on open-vocabulary state tracking tasks. This dataset is constructed by filtering and cleaning low-quality data from the original OpenPI dataset, and contains 13,800 state change records. The data is sourced from WikiHow, and has undergone a three-stage manual cleaning process to ensure data quality. OpenPI-C aims to address the accuracy and reliability issues of models when handling open-vocabulary state tracking, and holds significant applications particularly in mitigating model hallucinations.