MMChat
收藏资源简介:
MMChat是一个大规模的中文多模态对话数据集,包含120,840个过滤后的高质量对话和204,320张图片。该数据集从社交媒体中收集,特别关注图像引导对话中的稀疏性问题。数据集的创建过程包括从社交媒体收集原始对话,设计精细的数据过滤流程,以及手动标注部分对话以进一步过滤。MMChat的应用领域主要集中在开发更吸引人的对话系统,特别是在多模态上下文中的对话生成任务。
MMChat is a large-scale Chinese multimodal dialogue dataset comprising 120,840 filtered high-quality dialogues and 204,320 images. Collected from social media, this dataset specifically focuses on the sparsity issue in image-guided dialogues. The dataset creation process includes collecting raw dialogues from social media, designing a meticulous data filtering pipeline, and manually annotating a portion of dialogues for further screening. The main application scenarios of MMChat are centered on developing more engaging dialogue systems, particularly for dialogue generation tasks in multimodal contexts.

- 1MMChat: Multi-Modal Chat Dataset on Social Media阿里巴巴集团 · 2022年



