Next Generation Trustworthy Fraud Detection

Popular web applications, such as e-commerce, social networks and online ad auction, are providing valuable services to web users but have also been plagued by prevalent and diverse frauds. Many detection methodologies have been devised but detection trustworthiness is still one important and yet missing desideratum: a user will not trust a detector that has uncertain accuracy, can malfunction under unexpected situations, or can't explain its behaviors and interal working. Previous efforts mostly focused on detection accuracy, and our goal is to chart a path towards a more comprehensive definition of trustworthy detection, that consists of accuracy, transparency, and proactivity. To achieve the goal, we identify key challenges rooting at the specific settings of the above applications: the evolving nature and unexpectedness of the fraudsters' strategies, the ever-growing large amount of data, and the increasing complexity of effective detectors. We hope spark a large volume of research questions and solutions with respect to the above challenges.