Data Seminar
"Topic Modelling in Programmatic Advertising, and Mining Text Networks"The DABSS Seminar features a pair of related talks each month, one by a representative of a local company and one by a Duke researcher. This month we have: Title: Topic Modeling in Programmatic Advertising Speaker: Josh Howes, Senior Data Scientist, MaxPoint Abstract: I present an exploration of how MaxPoint uses web-scale topic modeling to better understand users, websites and behavioral trends in the high-speed world of algorithmic online advertising. AND Title: Mining Text Networks Speaker: Professor David Banks, Dept of Statistical Science, Duke University Abstract: Working with MaxPoint, we obtained the full text of all 1,509 U.S. political blog posts in 2012, as well as all links from one blog to another. This provided a dynamic text network, which we analyze through a combination of an evolving topic model and a dynamic network model, so that network structure improves topic discovery and the learned topics help to predict network change.