Adversarial attacks against graph neural networks (GNNs) through perturbations of the graph structure are increasingly common in social network tasks like rumor detection. Social media platforms capture diverse attack sequence samples through both machine and manual screening processes. Investigating effective ways to leverage these adversarial samples to bolster robustness is imperative. We enhance a maximum entropy inverse reinforcement learning method with a mixture-of-experts approach to address multi-source graph adversarial attacks. This method reconstructs the attack policy, integrating various attack models and offering feature-level explanations, subsequently generating additional adversarial samples to fortify the robustness of detection models. We devise precise sample guidance and a bidirectional update mechanism to mitigate the deviation resulting from imprecise feature representation and negative sampling within the expansive action space of social graphs, while also expediting policy learning. We take rumor detector as an example targeted GNNs model on real-world rumor datasets. By utilizing a small subset of samples generated by various graph adversarial attack methods, we reconstruct the attack policy, closely approximating the performance of the original attack method. We validate that samples generated by the learned policy enhance model robustness through adversarial training and data augmentation.