| The goal of viral marketing is that, by the virtue of mouth to mouth word spread, a small set of influential customers can influence greater number of customers. Influence maximization (IM) task is to discover such influential nodes (or customers) from a social network. Existing algorithms adopt Greedy based approaches, which assume only positive influence among users. But in real life network, such as trust network, one can also get negatively influenced.;In this research we propose a model, called T-GT model, considering both positive and negative influence. To solve IM under this model, a trust network where relationships among users are either `trust' or `distrust' is considered. We first compute positive and negative influence by mining frequent patterns of actions performed. Then using local search a new algorithm, called MineSeedLS, is proposed. Experimental results on real trust network shows that our approach outperforms Greedy based approach by almost 35%. |
