Multimodal Dataset for Sentiment Analysis and Classification

In the proposed dataset, videos are collected and downloaded from social media sites like Facebook, Instagram, YouTube, Twitter, Telegram, and video libraries that match requirements. The proposed dataset consists of 600 videos of different emotions like Neutral, Happy, Anger, Disgust, Fear, Sad, and surprise. The proposed dataset has videos of variable length from people of different ages and different nationalities and the audio in each clip has the voice of a single speaker who is seen in the film. The language of the videos is US English. It is labeled and balanced video dataset. This dataset can serve as valuable resource for various research purposes, including sentiment analysis, opinion mining, trend analysis, and social behavior studies, etc.

本研究提出的数据集从Facebook、Instagram、YouTube、Twitter、Telegram等社交媒体平台及符合筛选要求的视频库中收集并下载视频资源构建而成。该数据集共包含600段视频，涵盖中性（Neutral）、快乐（Happy）、愤怒（Anger）、厌恶（Disgust）、恐惧（Fear）、悲伤（Sad）以及惊讶（Surprise）7种情绪类别。数据集内的视频时长各异，拍摄主体涵盖不同年龄、不同国籍的人群，且每段视频的音频仅包含对应画面中单一说话者的语音。视频语言为美式英语（US English）。该数据集为带标注且类别均衡的视频数据集，可作为情感分析、观点挖掘、趋势分析以及社会行为研究等多种研究场景的宝贵资源。