Controllable uncertain opinion diffusion under confidence bound and unpredicted diffusion probability

• The diffusion probability is modified according to the distance between the opinions of interacted nodes.
• An optimization problem and suitable algorithm to show how to control the uncertain diffusion process.
• The performances of traditional influence maximization algorithms are often worse than that of random selection in this optimization problem.
• Reason of that the traditional influence maximization algorithms are not applicable to this problem.

The issues of modeling and analyzing diffusion in social networks have been extensively studied in the last few decades. Recently, many studies focus on uncertain diffusion process. The uncertainty of diffusion process means that the diffusion probability is unpredicted because of some complex factors. For instance, the variety of individuals’ opinions is an important factor that can cause uncertainty of diffusion probability. In detail, the difference between opinions can influence the diffusion probability, and then the evolution of opinions will cause the uncertainty of diffusion probability. It is known that controlling the diffusion process is important in the context of viral marketing and political propaganda. However, previous methods are hardly feasible to control the uncertain diffusion process of individual opinion. In this paper, we present suitable strategy to control this diffusion process based on the approximate estimation of the uncertain factors. We formulate a model in which the diffusion probability is influenced by the distance between opinions, and briefly discuss the properties of the diffusion model. Then, we present an optimization problem at the background of voting to show how to control this uncertain diffusion process. In detail, it is assumed that each individual can choose one of the two candidates or abstention based on his/her opinion. Then, we present strategy to set suitable initiators and their opinions so that the advantage of one candidate will be maximized at the end of diffusion. The results show that traditional influence maximization algorithms are not applicable to this problem, and our algorithm can achieve expected performance.