Ranking and Clustering in Signed and Weighted Bipartite Graphs
[Technical Report, Final Report]
ORTA DOGU TEKNIK UNIVERSITESI
Pagination or Media Count:
This project created and analyzed algorithms to cluster weighted graphs - ideas which could be applied to better understand radical subnetworks in social media. The project piggy packed with a larger US DoD effort and connected the PI with researchers working on DoDs Minerva Research Initiative at Arizona State University ASU. The project produced several conference papers as well as a journal article published jointly with the ASU team. The project successfully created several algorithms based on greedy heuristics to cluster bi-partite and tri-partite graphs. In addition to benchmarks on synthetic data, these algorithms were also tested on real-world Twitter data the goal to cluster UK Tweets from around to time of Brexit discussions to see if politicians, key words, and sentiment could be well identify by the clustering. While the algorithms did show promise, it remains challenging to directly compare these results to other existing clustering methods. Full details are found in the attached report as well as journal articles and conference papers therein. While AFOSR is currently not supporting follow-on efforts, the PI and his team plan to continue to improve their clustering techniques.
- Operations Research