PhD Scholarship - Building an Aussie Information Recommendation System You Can Trust

To receive quality recommendation services, online media platform users are commonly required to hand over their personal data, including personal attributes (e.g., age, gender, location, etc.) and previous interactions (e.g., information browsing history). Though recent research and industry practices have seen various privacy mechanisms, the degree of customizability is far from satisfactory as all users are restricted to a predefined set of sensitive attributes for protection. Hence, with the increasing public awareness and demand for online privacy, it is high time to revolutionize the privacy mechanisms of online media platforms by granting users full control over their sensitive information. Innovation. To meet users' varying privacy demands, this project put forward a graph-based user representation learning paradigm. This approach allows each user to disclose an incomplete yet distinct set of personal information while still learning expressive user representations from that data for downstream recommendation tasks.