Network structure engineering aims to intervene in the network structure to improve a specific underlying process and has applications in many research fields like biology, transportation, and network design. Here we address the problem of online social networks (OSNs) engineering by link addition through the friend suggestion for an efficient dissemination process. We propose an end-to-end machine learning framework called GCQL for optimizing or generating an effective network structure to improve the dissemination process in OSNs. We first propose a dissemination model called Complex-SII, combining two widely used dissemination models in OSNs, SIR and IC, and complex contagion. The GCQL next uses a reinforcement learning algorithm called Neural Fitted Q-Iteration for optimizing a network structure for the dissemination process and uses a graph convolution network called GCN for graph embedding and learning the important patterns and characteristics of network structure for efficient spreading. The GCQL is trained with small-sized graphs and uses the learning outcomes for medium/large-sized graphs. Our simulation results show that GCQL can generate relatively more efficient network structures than the greedy approach for small-sized networks. GCQL also uses these results to develop relatively more efficient networks than the greedy approach for medium/large-sized network structures at a much higher speed.