This paper proposes a novel model called Deep Interest Evolution Network (DIEN) for click-through rate (CTR) prediction. DIEN aims to capture the dynamic evolution of user interests over time, which is crucial for accurate CTR prediction in advertising systems. The model consists of two key modules: an interest extractor layer and an interest evolving layer. The interest extractor layer uses a GRU with auxiliary loss to capture temporal interests from user behavior sequences. The auxiliary loss helps in supervising the learning of hidden states, making them more expressive for interest representation. The interest evolving layer uses an attention-based GRU (AUGRU) to model the interest evolution process relative to the target item. This layer strengthens the influence of relevant interests and weakens the effect of irrelevant interests, leading to more accurate CTR predictions. DIEN significantly outperforms state-of-the-art solutions on both public and industrial datasets. It has been deployed in the display advertisement system of Taobao, achieving a 20.7% improvement in CTR. The model's ability to capture the dynamic evolution of user interests makes it particularly effective in e-commerce systems where user interests can change rapidly. DIEN's design addresses the limitations of previous models by incorporating auxiliary loss and attention mechanisms, leading to more accurate and effective CTR predictions. The model's performance is validated through extensive experiments and real-world deployment, demonstrating its effectiveness in practical advertising scenarios.This paper proposes a novel model called Deep Interest Evolution Network (DIEN) for click-through rate (CTR) prediction. DIEN aims to capture the dynamic evolution of user interests over time, which is crucial for accurate CTR prediction in advertising systems. The model consists of two key modules: an interest extractor layer and an interest evolving layer. The interest extractor layer uses a GRU with auxiliary loss to capture temporal interests from user behavior sequences. The auxiliary loss helps in supervising the learning of hidden states, making them more expressive for interest representation. The interest evolving layer uses an attention-based GRU (AUGRU) to model the interest evolution process relative to the target item. This layer strengthens the influence of relevant interests and weakens the effect of irrelevant interests, leading to more accurate CTR predictions. DIEN significantly outperforms state-of-the-art solutions on both public and industrial datasets. It has been deployed in the display advertisement system of Taobao, achieving a 20.7% improvement in CTR. The model's ability to capture the dynamic evolution of user interests makes it particularly effective in e-commerce systems where user interests can change rapidly. DIEN's design addresses the limitations of previous models by incorporating auxiliary loss and attention mechanisms, leading to more accurate and effective CTR predictions. The model's performance is validated through extensive experiments and real-world deployment, demonstrating its effectiveness in practical advertising scenarios.