SCS Team Wins Most Influential Paper Award at Data Mining Conference

CMU faculty members Leman Akoglu and Christos Faloutsos were two of three authors on a paper that received the Most Influential Paper Award at the 2020 Pacific-Asia Conference on Knowledge Discovery and Data Mining. Mary McGohon, now at Google, was the third author.

A 2010 paper by a trio of School of Computer Science researchers that described an algorithm for detecting spammers, faulty equipment, credit card fraud and other anomalous behavior won the Most Influential Paper Award at the 2020 Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD).

The paper was authored by Christos Faloutsos, the Fredkin Professor of Artificial Intelligence in the Computer Science Department (CSD), and two SCS students who subsequently received doctorates — Leman Akoglu, now Dean's Associate Professor in the Heinz College of Information Sciences; and Mary McGohon, site reliability engineering manager at Google. Both Faloutsos and Akoglu hold courtesy appointments with the Machine Learning Department; Akoglu is also affiliated with CSD.

The PAKDD is one of the top data mining conferences and its "most influential" award recognizes a paper published at the conference 10 years earlier that has had significant influence.

The trio's winning paper, "OddBall: Spotting Anomalies in Weighted Graphs," discussed a method for analyzing graphs to find nodes that could be considered strange. Such a technique could be used for a wide variety of applications, including detecting network intrusions, political campaign donation irregularities, and accounting inefficiencies or fraud. In social networks, it might detect unusual behaviors, such as adding friends indiscriminately in popularity contests.

Their algorithm sought out these anomalous nodes by looking not only at individual nodes, but also at their neighbors. These neighborhoods of nodes would forms spheres, or balls, hence the Oddball name for the algorithm. The algorithm proved to be fast and could function without supervision. It not only detected strangeness, but assigned an "outlierness" score to each node.

For More Information
Byron Spice | 412-268-9068 | bspice@cs.cmu.edu