Online Influence Maximization under Continuous Independent Cascade Model with Node-Edge-Level Feedback

We study the online influence maximization problem in social networks. We concentrate on solving two challenges in this paper. First, we work with continuous independent cascade model instead of independent cascade model. In the independent cascade model, the influence diffusion is limited and imprecise because an activated node can only attempt to impact a neighbor through a directed edge for once. Therefore we propose a continuous influence set, and combine this set with the independent cascade model to build a continuous independent cascade model, which realizes multiple activation and can be more customized to attract more targeted users. Second, we devise node-edge-level feedback instead of node-level feedback. In the node-level feedback, though the combined influence is relatively easy to observe, the exact edge which cause the influence seldom reveals. Therefore we use node-edge-level feedback to generate the source nodes which activate the active node. We improve the IMFB algorithm and propose the CIC_IMFB-NE algorithm. The CIC_IMFB-NE algorithm is more efficient than the existing online influence maximization algorithm. Our experiments demonstrate the excellence of CIC_IMFB-NE in terms of regret bound in real life.