Characterizing and using gullibility, competence, and reciprocity in a very fast and robust trust and distrust inference algorithm for weighted signed social networks

摘要

Predicting trust is a classic problem in social networks analysis. Furthermore, while most early approaches ignore distrust, recent works seem to consider it as important, if not more important, than trust itself. In this paper, we present a novel approach to predict both trust and distrust in Weighted Signed Social Networks very efficiently and in a satisfyingly accurate and robust way. Therefore allowing people to have healthier online presence and interactions.Being a local metric that does not rely on trust propagation, the proposed approach does not suffer from some serious limitations like trust decay, opinions conflict, path dependence, and time complexity. Moreover, our experiments on four real-world datasets show that, in addition to its simplicity and extensibility, this algorithm is robust to network sparsity, and provides satisfyingly accurate and very fast predictions.