Caser is a convolutional sequence embedding recommendation model designed for personalized top-N sequential recommendation. It models user behavior as a sequence of items and uses convolutional filters to capture both point-level and union-level sequential patterns, as well as skip behaviors. The model embeds the previous L items into an "image" in the latent space and uses convolutional filters to learn sequential patterns as local features of this image. This approach provides a unified and flexible network structure for capturing both general preferences and sequential patterns. Experiments on public datasets show that Caser consistently outperforms state-of-the-art sequential recommendation methods on various evaluation metrics. The model leverages convolutional neural networks to extract local features for image recognition and natural language processing, and it generalizes several existing methods within a single framework. Caser also allows for skip behaviors, which previous models failed to capture. The model's performance is evaluated on four datasets, and it outperforms baselines such as POP, BPR, FMC, FPMC, Fossil, and GRU4Rec. The results show that Caser is effective in capturing sequential patterns and improving recommendation accuracy. The model's components, including horizontal and vertical convolutional layers and personalization, contribute to its overall performance. The study highlights the importance of considering sequential information in recommendation systems and demonstrates the effectiveness of Caser in capturing both point-level and union-level sequential patterns.Caser is a convolutional sequence embedding recommendation model designed for personalized top-N sequential recommendation. It models user behavior as a sequence of items and uses convolutional filters to capture both point-level and union-level sequential patterns, as well as skip behaviors. The model embeds the previous L items into an "image" in the latent space and uses convolutional filters to learn sequential patterns as local features of this image. This approach provides a unified and flexible network structure for capturing both general preferences and sequential patterns. Experiments on public datasets show that Caser consistently outperforms state-of-the-art sequential recommendation methods on various evaluation metrics. The model leverages convolutional neural networks to extract local features for image recognition and natural language processing, and it generalizes several existing methods within a single framework. Caser also allows for skip behaviors, which previous models failed to capture. The model's performance is evaluated on four datasets, and it outperforms baselines such as POP, BPR, FMC, FPMC, Fossil, and GRU4Rec. The results show that Caser is effective in capturing sequential patterns and improving recommendation accuracy. The model's components, including horizontal and vertical convolutional layers and personalization, contribute to its overall performance. The study highlights the importance of considering sequential information in recommendation systems and demonstrates the effectiveness of Caser in capturing both point-level and union-level sequential patterns.