An Adversarial Graph Learning Approach for Networked Actor Comparative Assessment
Prof. Xi CHEN, Professor, Zhejiang University
Date: 26 February 2026 (Thursday)
Time: 10:00-11:30
Venue: E22-G008
Host: Prof. Yan LIN, Assistant Professor in Business Intelligence and Analytics
Abstract
The recent success of actors’ social influence evaluations on digital platforms has sparked interest in extending the progress to networked environments. In this study, we introduce a general causal assessment problem, Networked Actor Comparative Assessment (NACA), which estimates the differential social influence exerted by distinct actors on their connected individuals. A key yet understudied challenge in NACA is imbalanced network distributions in estimating individual treatment effects (ITEs): different actors differ in their social connections and peer features. The imbalances typically arise from actors’ heterogeneous tie establishment strategies, which can introduce systematic bias into causal assessments based on individual outcomes. We examine how these imbalances affect social influence assessment across actors and how to address these imbalances. Specifically, we propose Adversarial-based Graph Learning for Counterfactual Estimation (AGCE), a novel framework for social influence comparison across actors with observational networks. AGCE is a unified adversarial learning framework that incorporates graph-neural-network-based imbalances identification, counterfactual scenario simulation, and outcome prediction under balanced network conditions. Extensive experiments demonstrate that AGCE outperforms existing approaches in recovering causal differences and exhibits robustness across settings. We apply AGCE across multiple Twitter discussion domains to uncover the attitudinal social influence of introducing bots relative to humans. Moreover, we demonstrate that stakeholders can easily use AGCE to conduct networked counterfactual analysis and behavioral predictions. This complements existing policy tools for business actions such as social account early assessment and influencer selection.
Speaker
Chen Xi is currently a professor at the School of Management, Zhejiang University. His primary research interests include social media, social network analysis, and social commerce. He has led numerous national and provincial-level research projects, including Key Projects and Major Research Programs of the National Natural Science Foundation of China (NSFC), and General Programs of NSFC. His papers have been published or accepted in Information Systems Research (UTD24), Production and Operations Management (UTD24), INFORMS Journal on Computing (UTD24), Journal of Management Information Systems (FT50), Journal of the Association for Information Systems (ABS 4*), European Journal of Operational Research, as well as senior scholar basket journals in the information systems field such as Decision Support Systems and Information & Management. His research and teaching achievements have received multiple awards from provincial/ministerial-level institutions and international academic conferences. He serves as an associate editor for Information & Management and has served as track chairs or AEs for major IS conferences such as ICIS and PACIS.
All are welcome!
