M2H2

M2H2是一个专为多模态多方言印度语幽默识别设计的对话数据集，由印度理工学院巴特那分校计算机科学与工程系创建。该数据集包含6,191条来自流行电视系列'Shrimaan Shrimati Phir Se'的对话，每条对话均标注了幽默/非幽默标签，并涵盖了声音、视觉和文本三种模态。数据集的创建过程包括从YouTube下载视频文件和转录文件，并由专业人员进行转录和标注。M2H2数据集主要用于开发和评估多模态幽默识别系统，特别是在多语言和多方言环境中，以提升对话理解和交互的自然性和有效性。

M2H2 is a dialogue dataset dedicated to multimodal multi-dialect Hindi humor recognition, created by the Department of Computer Science and Engineering, Indian Institute of Technology Patna. This dataset includes 6,191 dialogues sourced from the popular television series *Shrimaan Shrimati Phir Se*. Each dialogue is annotated with binary humor/non-humor labels and encompasses three modalities: audio, visual, and text. The dataset construction process involves downloading video and transcription files from YouTube, followed by professional transcription and annotation by domain experts. The M2H2 dataset is primarily intended for developing and evaluating multimodal humor recognition systems, particularly in multilingual and multi-dialect settings, to improve the naturalness and efficacy of dialogue understanding and interaction.