Graphameleon: Relational Learning and Anomaly Detection on Web Navigation Traces Captured as Knowledge Graphs

Lionel Tailhardat, Benjamin Stach, Yoan Chabot, Raphaël Troncy

Abstract: User and Entity Behavior Analytics (UEBA) is key for managing security risks on information systems and comprehending user activities' impact on the network infrastructure. However, accessing network traffic and Web logs is challenging due to encryption or decentralized systems. Qualifying activities also requires contextualizing them according to the network's topology, as it determines potential exchanges and carries information about which services are used. This complexity hinders learning behavioral patterns when precise user action sequences are needed. We propose to tackle these challenges with Graphameleon, an open-source Web extension for capturing Web navigation traces. We model user activities in an RDF Knowledge Graph (KG), drawing from the UCO and NORIA-O ontologies. With this approach, we are able to distinguish analytics strategies implemented across different websites.