Topical unit classification using deep neural nets and probabilistic sampling

Abstract

Understanding topical units is important for improved human-computer interaction (HCI) as well as for a better understanding of human-human interaction. Here, we take the first steps towards topical unit recognition by creating a topical unit classifier based on the HuComTech multimodal database. We create this classifier by means of Deep Rectifier Neural Nets (DRN) and the Unweighted Average Recall (UAR) metric, applying the technique of probabilistic sampling. We demonstrate in several experiments that our proposed method attains a convincingly better performance than that using a support vector machine or a deep neural net by itself. We also experiment with the number of topical unit labels, and examine whether distinguishing between different types of topic changes based on the level of motivatedness is feasible in this framework.