LEGOBENCH

LEGOBENCH是由印度理工学院甘地分校LINGO实验室开发的数据集，用于评估科学排行榜生成系统。该数据集包含从arXiv获取的22年预印本提交数据和PapersWithCode门户上的11000多个机器学习排行榜。数据集设计了四种基于图和两种基于语言模型的排行榜生成任务配置，旨在通过这些任务配置全面评估系统的适应性和广泛性。LEGOBENCH的应用领域包括自动化排行榜生成，帮助研究人员跟踪最新模型，解决信息过载问题。

LEGOBENCH is a dataset developed by the LINGO Lab at the Indian Institute of Technology Gandhinagar for evaluating scientific leaderboard generation systems. This dataset includes 22 years of preprint submission data sourced from arXiv, alongside over 11,000 machine learning leaderboards from the PapersWithCode portal. The dataset provides four graph-based and two language model-based task configurations for leaderboard generation, aiming to comprehensively evaluate the adaptability and generalizability of target systems. Application domains of LEGOBENCH include automated leaderboard generation, assisting researchers in tracking state-of-the-art models and addressing the issue of information overload.