posted on 2022-12-01, 00:00authored byYingtong Dou
Recent years have witnessed the thriving of online services like social media, e-commerce, and e-finance. Those services facilitate our daily lives while breeding misbehaved actors like fraudsters and spammers to promote misinformation, gain monetary rewards, or reap end users’ privacy. Graph-based machine learning models have been playing a critical and irreplaceable role in modeling and detecting online misbehaviors. With the assumption that misbehaviors are different from massive regular behaviors, the graph models can leverage the relationship between users from a global perspective and reveal suspicious behaviors as anomalous nodes/edges/graphs. Despite the exceptional performance of graph models, their robustness against adversary and data concept drift is crucial to their sustainability. In my dissertation, I will present three published works and one working paper regarding robust graph learning against misbehaviors. The first work investigates the robustness of the graph-based spammer detection models using a reinforcement learning framework considering the spammers’ adversarial nature. The second work focuses on improving the robustness of the Graph Neural Networks (GNNs) under fraudster detection task by proposing a noisy neighbor filtering approach. The third work leverages GNNs to detect fake news on social media by encoding the news consumption preference of users. The last working paper focuses on sensing the misbehaviors from the graph concept drift perspective. All of my research works only use public-accessible data without IRB review.