CS 388, Fall 2018 Final Project: Link Detection in Political Networks

In this work we consider the task of link detection using a combination of natural language processing, network analysis and graph embedding methods over realworld politician networks. Networks for 219 Texas state and federal congressional members generated from a corpus of 50 thousand articles from various news sources in May 2015 are analyzed. We perform experiments comparing various baselines from each of the three methods and introduce a neural model which combines the representations from each baseline method. Factorization-, random walkand deep learning-based graph embedding methods are compared for effectiveness and error analysis is used to look at patterns against party affiliation, and congressional level ( state vs federal ). The network data set will be made available and interactive networks and article snippets are currently available.