Ranking Marginal Influencers in a Target-labeled Network

Using social networks for spreading marketing information is a commonly used strategy to help in quick adoption of innovations, retention of customers and for improving brand awareness. In many settings, the set of entities in the network who must be the targets of such an information spread are already known, either implicitly or explicitly. It would still be beneficial to route the information to them through a carefully chosen set of influencers in the network. We term networks where we have such vertices labeled as targeted recipients as targeted networks. For instance, in an online marketing channel of a fashion product, where vertices are tagged with 'fashion' as their preferred choice of online shopping, forms a targeted network. In such targeted networks, how to select a small subset of vertices that maximizes the influence over target nodes while simultaneously minimizing the non-target nodes which get the information (e.g., to reduce their spam, or in some cases, due to costs)? We term this as the problem of maximizing the marginal influence over target networks and propose an iterative algorithm to solve this problem. We present the results of our experiment with large information networks, derived from English Wikipedia graph, which show that the proposed algorithm effectively identifies influential nodes that help reach pages identified through queries/topics. Qualitative analysis of our results shows that we can generate a semantically meaningful ranking of query-specific influential nodes.