Collaborating as normal: Detecting systemic anomalies in your partner

Abstract

It is considered whether anomaly detection techniques might be used to determine potentially malicious behavior by service providers. Data mining techniques can be used to derive patterns of repeating behavior from logs of past interactions between service consumers and providers. Consumers may use these patterns to detect anomalous provider behavior, while providers may seek to adapt their behavior in ways that cannot be detected by the consumer. A challenge is deriving a behavioral model that is a sufficiently precise representation of the consumer-provider interactions. Behavioral norms, which model these patterns of behavior, are used to explore these issues in a on-line photograph sharing style service.