CC-Riddle

CC-Riddle是由北京大学和密歇根大学合作创建的中文谜语数据集，涵盖了大部分常用简体中文字符。该数据集通过网络爬虫、语言模型生成和人工筛选相结合的方式构建，旨在通过谜语描述和单个字符的解决方案来挑战语言模型对字符字形、常识和比喻语言的理解。CC-Riddle不仅用于评估语言模型处理比喻语言、常识和中文字符字形的能力，还希望推动更先进的中文和多语言语言模型的发展，为中文语言研究领域提供宝贵的辅助。CC-Riddle is a Chinese riddle dataset co-created by Peking University and the University of Michigan, covering most commonly used simplified Chinese characters. This dataset is constructed through a combination of web crawling, language model generation and manual filtering. It aims to challenge language models' understanding of Chinese character glyphs, common sense and figurative language via riddle descriptions and single-character solutions. CC-Riddle not only serves to evaluate the ability of language models to process figurative language, common sense and Chinese character glyphs, but also hopes to promote the development of more advanced Chinese and multilingual language models, providing valuable assistance for the field of Chinese language research.

搜集汇总

背景与挑战

背景概述

CC-Riddle是一个专注于汉字谜语的问答数据集，旨在支持汉字谜语理解和生成任务。该数据集来源于相关论文，提供生成和问答两种实验设置，并采用非商业许可协议（CC BY-NC-SA 4.0）。数据集包含Python和Shell脚本，便于用户复现实验。