University: Macquarie University
Academic supervisor:
Prof. Jian Yang
Industry partner: DFCRC
Academic Background and Work Experience
Venus Haghighi is currently a Postdoctoral Research Fellow in the Intelligent Computing Laboratory at Macquarie University. She has recently completed her PhD in Computer Science at Macquarie University under the supervision of Distinguished Professor Michael Sheng and Professor Jian Yang.
Her doctoral research focused on graph-based fraud detection, particularly addressing the challenges of identifying camouflaged malicious actors who deliberately mimic legitimate behaviours in multi-relational and heterophilic networks, where fraudulent and benign entities often appear structurally similar and are difficult to distinguish.
To overcome these challenges, she developed advanced solutions leveraging Graph Neural Networks (GNNs) to capture complex relational patterns, transformer-based hypergraph learning to model higher-order interactions beyond pairwise relationships, and LLM-augmented graph learning to integrate large language models for reasoning over graph-structured data.
Her work has been published in top-tier venues such as ICDM, CIKM, and IJCAI, and she has been recognized with several awards, including the HDR Research Rising Star Award issued by the School of Computing, the Digital Finance CRC PhD Top-Up Scholarship, and a prestigious Google Conference Travel Grant.
Her research interests include graph-based fraud detection, graph representation learning, hypergraph learning, large language models for graphs, and trustworthy AI.
Thesis Topic
Robust Graph-based Fraud Detection Against Camouflage
Fraud detection in online networks is increasingly challenged by camouflaged malicious actors who mimic legitimate behaviours. This thesis introduces adaptive graph learning frameworks to address the limitations of traditional GNNs in multi-relational and heterophilic settings. An adaptive aggregation mechanism captures diverse graph signals, while a relation-aware aggregation strategy is introduced to model cross-relation behavioural inconsistencies by capturing dependencies across multiple relation types. Building further, transformer-based hypergraph learning and LLM integration uncover complex fraud patterns. Experiments on real-world datasets demonstrate significant improvements in detecting sophisticated camouflaged fraudsters.
About DFCRC
DFCRC oversees and operates a 10-year, $180 million research program as a collaboration between industry partners, universities, and the Australian Government through the Cooperative Research Centres Program. Our mission is to develop and leverage the next transformation in financial markets – the digitisation of assets traded and exchanged directly on digital platforms.