MMChat中文社交媒体多模态聊天多轮对话数据集

MMChat是一个大规模社交媒体多模态多轮对话数据集，包含与图片相关联的中文对话，涵盖丰富的情感与社交场景。数据集通过多种策略保障对话质量，并提供图片与对话相关性的人工标注。其规模达数十万对话对，具有多模态、多轮交互和高质量语义关联等特点，适用于多模态对话系统、图文联合理解、情感分析等任务，推动社交场景下人机交互技术的研究与发展。

MMChat is a large-scale social media multimodal multi-turn dialogue dataset. It contains Chinese dialogues paired with associated images, covering a wide range of emotional and social scenarios. The dataset ensures high-quality dialogues through multiple strategies, and provides manual annotations for the relevance between images and their corresponding dialogues. With a scale of hundreds of thousands of dialogue pairs, it features multimodality, multi-turn interaction and high-quality semantic relevance. It is applicable to tasks such as multimodal dialogue systems, cross-modal image-text joint understanding, sentiment analysis and other relevant tasks, and promotes the research and development of human-computer interaction technologies in social scenarios.

搜集汇总

背景与挑战

背景概述

MMChat是一个大规模的中文社交媒体多模态多轮对话数据集，包含与图片相关联的对话，涵盖丰富的情感和社交场景。数据集通过多种策略保障对话质量，并提供图片与对话相关性的人工标注，具有多模态、多轮交互和高质量语义关联等特点，适用于自然语言处理和计算机视觉研究。