Hari Sundaram's research lies at the intersection of social network analysis and computational advertising—designing algorithms and building systems to understand and to influence collective behaviors in large social networks. His research has won awards from the ACM and IEEE societies.
He is focused on making fundamental methodological contributions while motivated by real-world applications. Some core contributions to network analysis include community discovery (Lin et. al., 2008; Lin et. al., 2011); network sampling ( Kumar 2016, de Choudhury et. al., 2010); influence maximization ( Sarkar 2016); detecting the onset of coordinated behavior (de Choudhury et. al., 2009) as well as large scale changes to network structure (Lin et. al., 2011). Some fun research questions on networked culture include: can a video game make us more productive (Nikkila et. al., 2013)? what makes a YouTube video interesting (de Choudhury et. al., 2009)?
Sundaram's current research is motivated by the challenge: how can we persuade millions of people to adopt behaviors that would be beneficial to them? Example behaviors include: leading healthy lifestyles; reducing individual energy consumption and greater civic engagement. The widespread adoption of these behaviors would lead to large scale societal benefits such as reduced healthcare costs, sustainability and a vibrant community. But, despite knowledge of benefits, many individuals do not adopt these behaviors.
To address the challenge of behavior change, one needs to innovate beyond addressing difficult questions in network analysis. In particular, one needs to incorporate ideas from different disciplines, including behavioral economics, psychology, information theory, advertising as well as stochastic control, as we develop new algorithms and build systems.
We cannot simply "cut & paste" ideas from other disciplines. Instead, we need to push ideas from these disciplines toward new directions now that we have real-time access to behavioral data of large populations, can perform sophisticated behavioral modeling and and since most individuals carry smartphones, we can deliver personalized messages. For example, we use ideas from stochastic control for real-time "nudging," synthesize personalized advertisements on the fly, use coding theory for distribution of persuasive messages in a social network, and finally insist on realistic assumptions—that individuals act under incomplete information, and are influenced by bounded resources—when developing models for behavior adoption.
Furthermore, we need to develop new theoretical frameworks to understand how behaviors spread in the physical world. These frameworks will be driven by field experiments to help us understand decision making, and should reveal fundamental limits: to what extent is behavior change possible? In other words, given a particular decision making model, a specific distribution of privacy profiles over the population (thus potentially limiting access to behavioral data), what target distributions of behavior over the population are achievable in a particular social network? Here is a map that summarizes my current research interests.
His group's current research projects include: network sampling and modeling (kumar and shah), the stability of social networks (agrawal, dev and geigle) network behavior summarization (krishnan, leung, liu and ho), persuasive messaging (xiao, zheng), evolution of individual behaviors (narang, chung and agarwal) and of beliefs (luber), game theoretic analysis of the diffusion of behavior in a social network (lee), experimental infrastructure (khandelwal), and privacy and data mining trade offs in the Internet of Things (de, choudhury, rao and wang).
He is also collaborating with several faculty at Illinois on research projects :internet of things (Kravets), persuasive messaging (Vargas ), crowd clustering (Parmeswaran), game theory and diffusion of behavior (Mehta), "nudging" and stochastic control of networks ( Langbort ), network "capacity" ( Varshney), network query models (Chang).