UC San Diego

Human Immuno-deficiency Virus(HIV) is one of the deadliest known viri to human-kind which when left untreated can lead to irreparable damages to the immune system. The body’s immune system cannot get rid of this virus and there is no medicine invented yet that can completely cure this disease. This though doesn’t mean that HIV isn’t treatable. Infected People can still lead better lives by detecting the infection early and employing better medical care. Therefore there is an imminent need in identifying HIV at-risk population and providing them with medical care. This can be achieved by adopting a methodical HIV risk assessment strategy that is fast and efficient.

Contemporary methods of HIV risk assessment have a long turn around time.It relies on static data from national census statistics and other surveys. Though in some cases this static data it is complemented with local data documented by the HIV clinics about their patient population, it only represents the specific subset of HIV at-risk individuals that are already seen at clinics for treatment or testing. Therefore the lack of early testing in many HIV at risk individuals hinders the ability to account for many people who are at risk. Hence the HIV population with the highest probability of infection transmission continue to remain undiscovered.

Our goal is to discover this population at an earlier stage as a near real-time intervention system to help doctors and researchers to respond to HIV risk effectively. This thesis is focused on the underpinnings of how we enable researcher's exploration of HIV epidemic and discuss the techniques that we used for it. It also elaborates on an integrated pipeline that is the backbone of this system. It is an amalgamation of components which perform natural language processing, supervised learning and network feature filtering of data publicly available, and confidentially collected from Twitter.

The ultimate goal of the developed infrastructure is to help clinicians to prune demographic information and social connections in the local population. We are going to discuss our novel contribution in task characterization and event extraction. We then discuss the integration of various additional components to PIRC-Net to support these additional features. Our integrated platforms allow clinicians to visualize this structure and identify patterns in online social media communication and provide an additional tool to inform targeted interventions.