Multimodal Sentiment Analysis

This talk describes a novel dataset consisting of videos collected from the social media website YouTube, which were annotated for sentiment polarity, and presents a method that integrates linguistic, audio, and visual features for the purpose of identifying sentiment in online videos.