Publication

Abstract : This paper investigates information diffusion on online social networks with an emphasis on the network evolution perspective. Unlike most previous works on information diffusion, the structural dynamics of the networks takes center-stage in this work. The proposed work identifies fundamental network structural properties that affect diffusion and generates a corpus of synthetic networks similar to a standard dynamic social network interactions dataset, using a random-rewiring strategy. The work uses machine learning models built for multi-target regression to model and predict the diffusion process on a network, given the structural evolution of a network. This modeling is done based on the data captured at three granularity levels: snapshot-wise, based on an evolution window of 100 snapshots, and an end-to-end evolution. The generated synthetic network corpus is validated using a graph kernel based on the Weisfeiler-Lehman graph isomorphism test for similarity with respect to the selected baseline dataset, and the multi-target regression models’ performance is analyzed based on standard metrics. The proposed model successfully learns from information flow patterns to reveal the complex relationships between network structure evolution and information propagation. The findings highlight the complex relationship between network structure and information movement in dynamic social networks. The multi-target regression models demonstrate robust predictive power, as indicated by the R 2 scores, which quantify the accuracy of the model’s ability to capture the underlying patterns of information diffusion in dynamic social networks.