Multimodal Features as a Novel Method for Cross-Cultural Studies

Abstract

The rise of media and new tools in computer science provide new approaches to study culture. We proposed a novel research method that leverages facial expression and language features from TV series to assist cross-cultural studies. We first compared the statistical results of the features and drew a set of conclusions that can be supported by a number of previous works in cultural studies. Then, we employed the multimodal features to train a classifier to recognize the country that a TV series belongs to. A linear SVM achieved a high average accuracy of 94.4%. We further interpreted the coefficients of the model and obtained new observations from the results. Our method can avoid the drawbacks of the existing cross-cultural research approaches, provide new ideas for media data analysis, and aid culture adaptive product design.