A Gamma-Poisson Mixture Topic Model for Short Text

Figure 1

The circles show the number of documents that contain the word “said” for different frequencies. The curve denotes predicted frequencies from a Poisson distribuition fitted to the data. Adapted from “Poisson mixtures” by Church and Gale [13]. Natural Language Engineering, 1 (2), 163–190. Copyright by Cambridge University Press 1995. Reproduced with permission.