Profit Maximization over Social Networks

TL;DR

This paper extends influence maximization models to explicitly include product adoption and profit considerations, proposing new algorithms that outperform existing methods in real-world social network datasets.

Contribution

It introduces a novel extension of the Linear Threshold model that incorporates prices and valuations, enabling profit maximization in social networks.

Findings

PAGE algorithm achieves highest expected profit

PAGE runs faster than other algorithms

Extended model maintains submodularity but loses monotonicity

Abstract

Influence maximization is the problem of finding a set of influential users in a social network such that the expected spread of influence under a certain propagation model is maximized. Much of the previous work has neglected the important distinction between social influence and actual product adoption. However, as recognized in the management science literature, an individual who gets influenced by social acquaintances may not necessarily adopt a product (or technology), due, e.g., to monetary concerns. In this work, we distinguish between influence and adoption by explicitly modeling the states of being influenced and of adopting a product. We extend the classical Linear Threshold (LT) model to incorporate prices and valuations, and factor them into users' decision-making process of adopting a product. We show that the expected profit function under our proposed model maintains submodularity under certain conditions, but no longer exhibits monotonicity, unlike the expected influence spread function. To maximize the expected profit under our extended LT model, we employ an unbudgeted greedy framework to propose three profit maximization algorithms. The results of our detailed experimental study on three real-world datasets demonstrate that of the three algorithms, \textsf{PAGE}, which assigns prices dynamically based on the profit potential of each candidate seed, has the best performance both in the expected profit achieved and in running time.