A fundamental understanding of human perception, sentiment, and impressions evoked from multimedia contents can benefit various applications starting from film-making over marketing up to video recommendations. In my research, I alleviate knowledge from human-generated data such as Social Media or the Web, to analyze the impression of multimedia contents. The first half of the talk introduces our method of estimating evoked emotions of Social media videos based on knowledge from their respective user-generated comments. The second half introduces research on estimating the perceived impression of a word or sound (i.e., the mental image it creates), based on analyzing Vision and Language data. Through these researches, we can get a step closer to fully understanding human impressions of multimedia contents.
Invited talk at Interdisciplinary Workshop on Analysis of TV content (ERC Demoseries)