Fraud detection based on graph learning in social networks

PhD

Fraud detection based on graph learning in social networks

Outline

Research has shown that social networks are the main source of obtaining confidential information to carry out fraudulent activities, such as laundering transactions and false information spreading. As large amounts of information are made public, how to conduct fraud detection to increase the security and privacy of social networks becomes an important problem. However, it is difficult to detect fraud at the age of big data. On one hand, the small number of fraudulent samples leads to existing supervised learning methods useless. On the other hand, it is difficult to distinguish different types of fraud under the condition of multi-source data.

During this study towards the doctoral degree, the student will focus on studying the following three questions in response to the above difficulties and challenges:

1.    How to detect fraud with few labelled fraudulent samples?

2.    How to distinguish different types of fraud with multi-source data?

3.    How to explain the fraud detection results?

To address these questions, this project will develop a suite of innovative solutions taking advantage of the graph learning framework. The outcomes of this research will lead to several high-quality publications in leading journals and conferences.

Supervisory team

Principal Supervisor: A/Prof Feng Xia

Co-supervisor:

Dr Muhammad Usman

Dr Taiwo Oseni