MMDialog
收藏资源简介:
MMDialog是一个大规模的多轮对话数据集,旨在促进多模态开放领域对话的研究。该数据集由北京大学和微软公司合作创建,包含1.08百万个真实世界的对话和1.53百万个独特的图像,覆盖4,184个话题。数据集的创建过程中,通过精心设计的数据过滤和后处理步骤,确保了数据的质量和多样性。MMDialog的应用领域广泛,旨在解决多模态对话系统中的生成和检索任务,为构建更吸引人的多模态对话代理提供支持。
MMDialog is a large-scale multi-turn dialogue dataset dedicated to advancing research in open-domain multimodal dialogue. It was co-created by Peking University and Microsoft Corporation, containing 1.08 million real-world dialogues and 1.53 million unique images across 4,184 topics. During its curation, meticulously designed data filtering and post-processing procedures were implemented to ensure the dataset's quality and diversity. MMDialog has broad application scenarios, aiming to address the generation and retrieval tasks in multimodal dialogue systems and provide support for building more engaging multimodal dialogue agents.




