Abstract:

In online social networks, the cascading of sensitive data such as personal material and gossip is a serious problem. The dispersion of social network users is constrained by one technique to minimize cascade of sensible information. Nevertheless, the measures restricting diffusion restrict the transmission of sensitive information, which leads to the unpleasant experience of users. In order to address this problem, we examine the problem of minimising sensitive dissemination while preserving the distribution of non-sensitive information. This problem is defined as a restricted minimization problem, where the intention of preserving non-sensitive dissemination is characterised as a limitation. We examine the problem with the well-known network of all users and the semi-known network, in which partial users remain unaware of their dispersion ability. When modelling the delicate spread of information as a reward for a bandit, we use the bandit framework to create solutions with polynomial complexity in both cases together. Moreover, it is impossible to measure the information diffusion size of the algorithm design due to the unknown diffusion capacity of the semi-known network.. For this matter we propose to learn in real time the unaccounted capacity of diffusion of the diffusion process and then to adjust to the diffusion restrictions, using the bandit framework, based on the knowledge of diffusion abilities. Large studies in actual and simulated data sets show that our methods are able to successfully limit the dissemination of sensitive data, while losses of non-sensitive information are 40 percent less compared to four baseline techniques.

Keywords: