Solving the Influence Maximization Problem Using a Genetic Deep Reinforcement Learning Approach

The influence maximization (IM) problem is currently a rising research hotspot in the complex network field. Nodes with the best information dissemination effect are expected to be selected. At present, algorithms and diffusive models have been developed, and the IM problem can be solved as a continuous parameter optimization one. Although encouraging results can be obtained, the existing studies have not considered the impact by different coding methods of the seed determination process. Focusing on this deficiency, this paper proposes an algorithm framework combined the graph embedding method with the deep reinforcement learning to iteratively search for competitive seeds under a genetic framework, termed SDNE-GDRL. Both optimal and structural information are considered to guarantee the search ability. Experiments have been conducted on several networks with different sizes, which reveal that the proposed algorithm shows superiority over existing approaches.