Logo image
Back
Towards community-based influence spread prediction (CIP) for edge changes in large-scale dynamic social networks
Conference proceeding

Towards community-based influence spread prediction (CIP) for edge changes in large-scale dynamic social networks

Pranav Pamidighantam, Vairavan Murugappan, Suresh Subramanian and Eunice E. Santos
2025 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), pp 265-274
03 Jun 2025
DOI: https://doi.org/10.1109/IPDPSW66978.2025.00050

Abstract

Accuracy Community detection Community-based influence Conferences Distributed processing Dynamic Influence Prediction Dynamic Networks Heuristic algorithms Image edge detection Influence learning Influence maximization Influence spread Information diffusion Information diffusion models Large-scale networks Prediction algorithms Predictive models Social networks Susceptibility to influence Training Machine Learning
Learning, predicting, and understanding influence spread and information diffusion is a central problem in social network analysis. The dynamic nature of real-world social networks with continuously evolving connections presents a significant challenge to efficiently predict influence spread. In our prior work, we introduced a novel community-based influence spread prediction (CIP) methodology that leverages susceptibility to influence, along with inherent network traits and community structure, to understand and predict influence spread. In this paper, we extend CIP and present an efficient approach for predicting influence spread in large-scale, dynamic social networks, specifically during edge changes. Influence spread in networks is often driven by within-community behavior. Hence, we focus on simulating, training models, and predicting this within-community spread by identifying representative communities and generalizing the learned spread models. Preliminary results demonstrate that the proposed approach predicts influence spread within most communities with high accuracy. We show there are cases in which there is no substantial difference in accuracy between retraining and using models trained on original networks. Beyond within-community spread, we also test CIP's ability to predict whole-graph spread. We find that as edges are added (uniformly randomly), the difference between spread achieved using an influence maximization algorithm and the proposed model increases. Based on these results we also present future research directions to address extending the accurate within-community predictions to whole-graph spread predictions.

Metrics

1 Record Views

Details

Logo image