Indirect Influence Manipulation with Partial Observability

Abstract

The propagation of concepts through a population of agents can be modelled as a cascade of influence spread from an initial set of individuals. In real-world environments there may be many concepts spreading and interacting, and we may not be able to directly control the target concept we wish to manipulate, requiring indirect manipulation through a secondary controllable concept. Previous work on influence spread typically assumes that we have full knowledge of a network, which may not be the case. In this paper, we investigate indirect influence manipulation when we can only observe a sample of the full network. We propose a heuristic, known as Target Degree, for selecting seed nodes for a secondary controllable concept that uses the limited information available in a partially observable environment to indirectly manipulate the target concept. Target degree is shown to be effective in synthetic small-world networks and in real-world networks when the controllable concept is introduced after the target concept.