1 Introduction
Methods of data analysis such as visualization, automatic classification, and structure discovery in large collections of data gain a growing practical importance in many domains. Generative models are frequently used as tools for discovering latent structure or latent causes in high dimensional observable data. Areas of applications include data mining, telecommunications, bioinformatics, fraud detection, information retrieval, and marketing analysis. In particular, the modeling algorithm introduced in this paper, is well-suited for unsupervised discovery of semantic structure from text databases.