UC San Diego

The world continues to battle with HIV, one of the major public health issues. It has no known cure yet and has claimed several million lives so far. Over a period of time, there has been tremendous progress in intervention methods to prevent the spread of this disease. Traditional intervention schemes have however been faced with several challenges. An important challenge is the lack of a real-time radar to address the most at-risk communities at a given time and location. Another challenge is related to limited HIV prevention resources. Such challenges makes it important to progress towards a more targeted, evidence based, and data-driven intervention method. Along with being data-driven, such intervention also needs to be effective, scalable, and tailored as per the distinct needs of communities.

The primary goal of this thesis is to unveil HIV transmission networks with the help of information gathered from digital social media footprint. Recent research in this area suggests the feasibility of using Twitter as a platform to uncover HIV at-risk behaviors among communities. Our research further explores this by trying to understand how likely a person may be at the risk of acquiring or transmitting HIV based on the nature of the content shared or consumed via Twitter. We focus on the quality of the data gathered, as well as its temporal and spatial dimensions to help with a more real-time risk analysis.

To begin with, we look only at the text-based information shared by users to design a model for HIV risk prediction. We move on to improving the risk analysis model by exploiting social relationships from network induced user-user connections in addition to the text-based information. Towards the end we discuss how our model can be used to inform the current intervention methods, followed by areas for improvement and future directions to further refine our techniques.