Evolutionary Algorithm For Seed Selection In Social Influence Process

Nowadays, in the world of limited attention, the techniques that maximize the spread of social influence are more than welcomed. Companies try to maximize their profits on sales by providing customers with free samples believing in the power of word-of-mouth marketing, governments and non-governmental organizations often want to introduce positive changes in the society by appropriately selecting individuals or election candidates want to spend least budget yet still win the election. In this work we propose the use of evolutionary algorithm as a mean for selecting seeds in social networks. By framing the problem as genetic algorithm challenge we show that it is possible to outperform well-known greedy algorithm in the problem of influence maximization for the linear threshold model in both: quality (up to 16% better) and efficiency (up to 35 times faster). We implemented these two algorithms by using GPGPU approach showing that also the evolutionary algorithm can benefit from GPU acceleration making it efficient and scaling better than the greedy algorithm. As the experiments conducted by using three real world datasets reveal, the evolutionary approach proposed in this paper outperforms the greedy algorithm in terms of the outcome and it also scales much better than the greedy algorithm when the network size is increasing. The only drawback in the GPGPU approach so far is the maximum size of the network that can be processed - it is limited by the memory of the GPU card. We believe that by showing the superiority of the evolutionary approach over the greedy algorithm, we will motivate the scientific community to look for an idea to overcome this limitation of the GPU approach - we also suggest one of the possible paths to explore. Since the proposed approach is based only on topological features of the network, not on the attributes of nodes, the applications of it are broader than the ones that are dataset-specific.