Role Recognition in Radio Programs using Social Affiliation Networks and Mixtures of Discrete Distributions: an Approach Inspired by Social Cognition

This paper presents an approach for the recognition of the roles played by speakers participating in radio programs. The approach is inspired by social cognition, i.e. by the way humans make sense of people they do not know, and it includes unsupervised speaker clustering performed with Hidden Markov Models, Social Network Analysis and Mixtures of Bernoulli and Multinomial Distributions. The experiments are performed over two corpora of radio programs for a total of around 45 hours of material. The results show that more than 80 percent of the data time can be labeled correctly in terms of role.