Deep User Interest Mining Network Based on Attention Mechanism and Deep Learning

Click-through rate prediction plays an important role in recommendation systems and online advertising. Most CTR prediction models focus more on innovative research on feature interaction, but ignore the mining and application of user behavior sequences. There are many shortcomings and shortcomings in the in-depth mining of user interests, especially in the use of the user's historical behavior sequence characteristics, which can still be optimized and improved. In this article, we propose a new model: User Interest Mining Network (UIMN). This model innovatively designs the User Behavior Interest Network (UBIN), which is used to deeply mine user interests, and at the same time innovatively introduce User-Item Relational Network (UIRN). The model processed the user's behavior sequence in long and short term, and considered the influence of the target item on the sequence, and finally got the updated behavior sequence. In addition, the model introduces the relationship between users and target item, and combines DNN and LSTM networks to improve the accuracy of estimation. UIMN con-ducted experiments on Amazon commodity dataset and MovieLens dataset, and the AUC value of the model increased by 1.6% to 2.4% compared with the benchmark model, which proved the effectiveness of the model.