In recent years, the sequential recommendation achieves excellent results. But it also meets many challenges. For example, when we add adversarial perturbations to the input, the model’s performance might be weakened. To solve this problem, we propose a novel model named APRSR, Adversarial Personalized Ranking Modeling for Sequential Recommendation in short. APRSR model is designed based on a self-attention sequential recommendation model by adding adversarial perturbations. First, APRSR can learn local representation and the global representation respectively, then it can get the final representation by balancing the local representation and the global representation. In the meantime, it considers the influence of candidate items on user’s intent. To enhance the robustness and learn more expressive features, APRSR utilizes the idea of the adversarial matrix factorization which can generate the adversarial perturbations. Extensive experiments on five public real-world datasets demonstrate the effectiveness of APRSR and outperform those of other state-of-the-art models.