Optimizing Interventions for Changing HIV Risk Behaviors via Temporal Link Prediction in MSM Social Networks

Description

In Hong Kong and Mainland China, men who have sex with men (MSM) are experiencing a serious human immunodeficiency virus (HIV) epidemic with an exponentially growing rate. Additional intervention efforts are needed to prevent the further transmission of HIV among MSM.  Network interventions achieve behavioral changes by inducing social influence and peer-pressure in social networks. Common HIV network interventions utilize the social capital of individuals to influence others in the community to alter their HIV risk behaviors (e.g. unprotected sex). However, existing network interventions are limited in several ways. First, existing studies usually focus on one type of behavioral changes, while ignoring the multiplexity nature of risk behaviors. Second, existing methods for selecting seed individuals either ignore the overall network structure, or only reply on predefined topological properties, thus are usually not optimal. Third, existing network interventions are based on observed social networks without considering future changes in the network structure. Such reactive interventions are not optimal when the network structure changes. Given these limitations, it is needed to develop predictive analytics tools to optimize the HIV network interventions that involve multiple types of behavioral changes in the dynamic social networks of MSM in a proactive manner. There are three main methodological challenges in incorporating predictive analytics into decision making for HIV network interventions, including (a) extracting the multiaspect latent features in the complex and dynamic social networking behaviors, (b) integrating the latent features with network structure and explicit features for temporal link prediction, and (c) utilizing both the observed and predicted social network structures to facilitate the search for optimal intervention strategies. To address the aforementioned challenges, this project aims to develop a predictive decision analytics system to enable proactive HIV network interventions for MSM. More specifically, the proposed system consists of a novel tensor factorization-embedded deep learning model that predicts social relations among individuals in an MSM community recursively, and a novel network-enhanced optimization algorithm that identifies the optimal subset of seed individuals to induce multiple types of behavioral changes in the MSM community. The development of the proposed analytics methods is based on data from existing MSM cohorts. To evaluate the effectiveness of the proposed system, the team will further recruit cohorts in Hong Kong and Guangzhou for real-world intervention studies. This project will generate data-driven insights and analytics tools that lead to proactive and costeffective HIV network intervention programs for MSM.