AIRQA-REAL
收藏资源简介:
AIRQA-REAL是一个针对学术论文的问答数据集,包含553个样本和18个实例特定的评估指标。该数据集由上海交通大学人工智能教育部重点实验室、苏州语言计算江苏省重点实验室和苏州AISpeech有限公司共同创建。数据集内容主要来源于2023年和2024年发表的AI论文,旨在解决学术论文中关键细节的提取问题。数据集创建过程中,研究人员采用了多视图分块和模式化解析的方法,将半结构化的PDF内容组织成关系数据库和向量存储,从而实现了LLM代理的迭代收集上下文直到生成答案的能力。
AIRQA-REAL is an academic paper-oriented question answering dataset consisting of 553 samples and 18 instance-specific evaluation metrics. It was jointly developed by the Key Laboratory of Artificial Intelligence of the Ministry of Education, Shanghai Jiao Tong University, the Jiangsu Key Laboratory of Language Computing, Soochow University, and Suzhou AISpeech Co., Ltd. The dataset content is primarily sourced from AI papers published in 2023 and 2024, aiming to address the challenge of extracting key details from academic papers. During the dataset construction phase, researchers adopted multi-view chunking and schema-based parsing methods to organize semi-structured PDF content into relational databases and vector stores, thereby endowing LLM Agents with the ability to iteratively gather context until generating final answers.




