Competition-based generalized self-profit maximization in dual-attribute network

Abstract

Profit Maximization in social advertising aims at selecting some users of social network as initial adopters and information sources to trigger the spread of promotion information such that the profit generated by all the adopters reaches maximum when the dissemination terminates. A lot of related works mainly study this problem under three assumptions: pure network, single product and one-dimension diffusion model. However, in real advertising activities conducted in social networks, advertisement of competitive products may spread almost at the same time. And there are many factors that can influence the probability with which a potential consumer makes an adoption. For the purpose of approximating real social advertising, we propose the Dual-Attribute Compete (DAC) model where attributes of both potential consumers and competitive products are taken into consideration and the information about competitive products can spread simultaneously. Therefore, it can capture not only the competition between different products but also the reaction of potential consumers to products. Under DAC model, we study the Competition-based Generalized Self-profit Maximization (CGSM) problem whose purpose is selecting at most k individuals to form an optimal seed set as the source of information diffusion to maximize the profit related to adopters. Given that the objective function of CGSM problem is generally nonsubmodular, we design R-CGSM algorithm to tackle it. Based on the analysis of martingale and the concept of Shapley value, this algorithm uses sandwich method to get a pretty good solution of CGSM problem. We evaluate the R-CGSM algorithm by conducting experiments on four different data sets representing a synthetic network and three social networks in real world, respectively. Results of experiments validate the effectiveness and accuracy of R-CGSM algorithm.