Online Influence Maximization with Node-Level Feedback Using Standard Offline Oracles

We study the online influence maximization (OIM) problem in social networks, where multiple rounds learner repeatedly chooses seed nodes to generate cascades, observes cascade feedback, and gradually learns best seeds that largest cascade. focus on two major challenges this paper. First, we work with node-level feedback instead of edge-level feedback. The reveals all edges pass through information a cascade, whereas only activated timestamps. is arguably more realistic since practice it relatively easy observe who influenced but very difficult from which relationship (edge) comes. Second, use standard offline oracles pair-oracles. To compute good set for next round, an pair-oracle finds parameters within confidence region simultaneously, such oracle due combinatorial core OIM problem. So how given edge as input. In paper, resolve these famous independent (IC) diffusion model. past research achieves while present first optimal regret algorithm For challenge above, apply novel adaptation maximum likelihood estimation (MLE) approach learn graph its (a ellipsoid). second challenge, adjust update procedure dissect ellipsoid into intervals each parameter, so enough.