Influence Maximization Using Observational Data

Mohith Murlidhar, Purdue University

Abstract

Influence maximization is the problem of finding a subset of nodes (seed nodes) in a network, such that by incentivizing these set of nodes, one maximizes the expected spread of influence in the network. This is an NP hard problem for which in the past, greedy algorithms and heuristics have been proposed for solving the problem. Most of these algorithms on this topic have focused on using information about the underlying graph and the edge probabilities, while running expensive controlled experiment simulations to find a solution to the influence maximization problem. These algorithms also accounted only for direct (or indirect) influence of the seed nodes and did not consider the possibility of spontaneous infections i.e. nodes that are already active before being incentivized. Proposed here is an approach to finding a solution to the influence maximization problem using observational data. To work with observational data, Judea Pearl's Causal Calculus is used, a methodology that employs Bayesian Networks to infer causal relationships from data. However, working with Bayesian Networks restricts the scope of the problem to directed acyclic graphs. The performance of the proposed algorithm is validated using simulated data and the performance of it is compared to existing, well established, greedy and efficient algorithms.

Degree

M.S.I.E.

Advisors

Quinn, Purdue University.

Subject Area

Mathematics|Operations research|Computer science

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS