A Hierarchical Attention Model for CTR Prediction Based on User Interest

Abstract

The prediction of click-through rate is a challenging problem in the aspect of online advertising. Recently, researchers have proposed deep learning-based models that follow a similar embedding and multilayer perceptron paradigm. Although encouraging successes have been obtained, the importance of capturing the latent user interest behind user behavior data was ignored by most of the methods, which has the potential to effectively learn the feature interactions. In this article, we propose an attentive-deep-interest-based model to fill these gaps. Specifically, we capture the interest sequence in the interest extractor layer, and the auxiliary losses are employed to produce the interest state with deep supervision. First, we use the bidirectional long short-term memory network to model the dependence between behaviors. Next, an interest evolving layer is proposed to extract the interest evolving process that is related to the target. Then, the model learns highly nonlinear interactions of features based on stack autoencoders. An experiment is conducted using four real-world datasets. The experimental results show that the proposed model achieves 1.8% improvement in the Amazon datasets than the existing state-of-the-art models.