A Hierarchical Bayesian Model for Topic Segmentation

Many streams of real-world data, such as conversations or body movements, consist of relatively coherent segments, each characterized by particular topics or controllers. Making sense of these data requires simultaneously segmenting the sequences and inferring the structure of the segments. We present a hierarchical Bayesian model that can be used to break a sequence of utterances or movements into segments with different distributions over topics or controllers. We apply this model to a database of meetings, showing that its unsupervised segmentation is competitive with other approaches, and a database of human hand movements, revealing some of the controllers for motions of the hand.