November 30, 2017
All Day
209 W Eighteenth Ave (EA), Room 170
Title
Dynamic Text Networks
Speaker
David Banks, Duke University
Abstract
There are many examples of dynamic text networks, such as the Wikipedia and citation networks. This research combines models for dynamic networks with topic models in the context of U.S. political blogs. The key idea is that if two bloggers are linking to each other, then they are discussing the same topic, so the network structure should inform topic discovery. Similarly, if two bloggers are interested in the same topic, they are more likely to link, so the topic mining should improve inference on network dynamics. This leads to new methods for community detection and interpretable insight into topic evolution.