Time Advertising Inc.
Mining Political Blogs With Network Based Topic Models
We develop a Network Based Topic Model (NBTM), which integrates a Random Graph model with the Latent Dirichlet Allocation (LDA) model. The NBTM assumes that the topic proportion of a document has a fixed variance across the document corpus with author differences treated as random effects. It also assumes that the links between documents are binary variables whose probabilities depend upon the author random effects. We fit the model to political blog posts during the calendar year 2012 that mention Trayvon Martin. This paper presents the topic extraction results and posterior prediction results for hidden links within the blogosphere.