Data-Driven Influence Learning in Social Networks

摘要

How to model the influence diffusion process accurately is an open issue that it has attracted a lot of researchers in the field of social network analysis. The existing researches assume they have already owned the social graphs with edges labeled with the influence probability. However, the question of how to obtain these probability from social networks has been largely ignored. Thus, it is interesting to address the problem of how to model the influence diffusion based on the data of social graphs and action logs. This is the main problem we addressed in this paper, and our purpose is to solve the problem of seeds detection via the data-driven influence probability calculation. We consider the influence probability can be viewed as two parts of the influence strength and the influence threshold. For learning the influence probability, we propose a novel Data-driven Influence Learning (DIL) algorithm including three stages. The experimental results illustrate our algorithm performs better than other baselines in various datasets. In addition, our algorithm enables us to detect the seed sets in large social networks.