Cultural Differences Demonstrated by TV Series: A Cross-Cultural Analysis of Multimodal Features

Abstract

TV series is one of the most popular entertainment media globally and is a representation of popular culture. It reflects a specific group's daily life culture and certain characteristics of society such as social norms. This paper improved and verified a new cross-cultural analysis method by analyzing facial expressions, original text features and audio features extracted from TV series datasets. We adopted the TV series from America, Japan, and Korea and extracted the textual features from the original text database rather than the translated one. We added the emotional frequency of text and part-of-speech frequency in the text modality. The emotional frequency of facial expressions and text were combined to explore the relation between nonverbal and verbal expressions. In addition, the feasibility of using audio features to further extend the new cross-cultural analysis method were explored. Overall, 1656 features extracted from 90 TV dramas were analyzed, including 42 facial features, 32 text features and 1582 audio features. The statistical results of the feature comparisons revealed the similarities and differences between the three countries and agreed with many existing theories, which resulted in traditional cross-cultural studies. Machine learning models of random forest and support vector machine were used for feature selection and classification to enhance the understanding of important features and conduct country classification.