2155889636

Applications:
Computational Linguistics, Information Systems, Networks, Social Sciences
Methodologies:
Causal Inference, Deep Learning, Econometrics, Machine Learning, Natural Language Processing, Statistical Inference

Paramveer Dhillon

Assistant Professor

School of Information

My research interests span topics in Statistical Machine Learning, Computational Social Science, Natural Language Processing, and Field/Digital Experiments. Substantively, I am interested in understanding the impact of internet technologies on users by empirically studying their interactions with such systems. The research questions that I study are of both predictive as well as causal nature and I examine them by using data from text/natural language and social network domains.

The same network is displayed, parameterized by four different models of the distribution of influence and susceptibility over nodes, characterized by four types of nodes: low influence and low susceptibility nodes, high influence and low susceptibility nodes, high influence and high susceptibility nodes and low influence and high susceptibility nodes. The optimal seed nodes selected under each model are outlined in green. a, Baseline IC and LT models for which propagation properties are specified as constant (top) and the inverse of node degree (bottom), respectively. b, Baseline IC and LT models for which propagation properties are specified according to the assortative influence, assortative susceptibility, substitute influence–susceptibility (AAS) model. c–e, The same information as in b, but for the assortative influence, disassortative susceptibility, substitute influence–susceptibility (ADS; c), disassortative influence, disassortative susceptibility, substitute influence–susceptibility (DDS; d) and disassortative influence, assortative susceptibility, substitute influence–susceptibility (DAS; e) empirical influence models. Distributions of the frequency of the four types of nodes with different influence and susceptibility characterizations are displayed underneath each graph or model. Seed sets differ substantially across different parameterizations of the graph, implying vastly different influence maximization results for the different models of influence and susceptibility. **This figure is taken from the paper: “Social influence maximization under empirical influence models” by (Aral and Dhillon, Nature Human Behaviour 2018).**