Sparsification of influence networks

Michael Mathioudakis, Francesco Bonchi, Carlos Castillo, Aristides Gionis, Antti Ukkonen
2011 Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining - KDD '11  
We present Spine, an efficient algorithm for finding the "backbone" of an influence network. Given a social graph and a log of past propagations, we build an instance of the independent-cascade model that describes the propagations. We aim at reducing the complexity of that model, while preserving most of its accuracy in describing the data. We show that the problem is inapproximable and we present an optimal, dynamic-programming algorithm, whose search space, albeit exponential, is typically
more » ... ch smaller than that of the brute force, exhaustive-search approach. Seeking a practical, scalable approach to sparsification, we devise Spine, a greedy, efficient algorithm with practically little compromise in quality. We claim that sparsification is a fundamental datareduction operation with many applications, ranging from visualization to exploratory and descriptive data analysis. As a proof of concept, we use Spine on real-world datasets, revealing the backbone of their influence-propagation networks. Moreover, we apply Spine as a pre-processing step for the influence-maximization problem, showing that computations on sparsified models give up little accuracy, but yield significant improvements in terms of scalability.
doi:10.1145/2020408.2020492 dblp:conf/kdd/MathioudakisBCGU11 fatcat:v35756h7xndrrgsosrkp63qgha