Explicit and Implicit Feature Interaction Based on Attention Networks for Click-Through Rate Prediction

Click-through rate (CTR) prediction, which aims to estimate the probability of a user to click on a given item, plays a critical role in both recommender systems and online advertising. Recently, a number of deep neural network-based approaches have been developed in an implicit and bit-wise interaction, which the order of interaction is inconclusive. In this work, a novel model called eXtreme Deep Attention Interaction Network (xAtInt) is proposed to capture meaningful combination features explicitly. On the one hand, xAtInt learn meaningful feature interaction via attention mechanism; on the other hand, it is capable of finding arbitrary degree interaction at a vector-wise. Furthermore, to improve performance, we integrate DNN with xAtInt learning explicit and implicit feature interactions. Extensive experiments are conducted on real-world dataset. The results show that our model present good performance compared existing state-of-the-art approaches.