Adversarial Training, as a paradigm in deep learning, has made significant progress in improving model performance. However, in sequence recommendation tasks, facing the challenges of handling sequential data and capturing long-range dependencies, more powerful modeling tools are still needed. To address this issue, a novel approach named Adversarial Training and Contrastive Learning with Bidirectional Transformers for Sequence Recommendation (ACBiT) is proposed, which cleverly integrates adversarial training and contrastive learning, innovatively training bidirectional Transformers. Specifically, employing reinforcement learning and policy gradient ideas, through the adversarial training framework of generator and discriminator, the aim is to enhance the model’s performance in sequence generation tasks. The generator architecture adopts bidirectional Transformers, focusing on modeling bidirectional information of sequential data. The discriminator, on the other hand, utilizes a CNN structure, concentrating on capturing the local features of sequences. During the Monte Carlo sampling process, a contrastive learning task is introduced to provide additional training signals for the generator, thereby enhancing the model’s understanding and recommendation performance. Experimental results demonstrate that the proposed method achieves significant improvements compared to traditional approaches in sequence recommendation tasks. This study provides an effective way to enhance the performance of deep learning methods for recommendation systems, offering important guidance for constructing more intelligent, accurate, and personalized recommendation systems.