NAUI : Neural Attentive User Interest Model for Cross-Domain CTR Prediction

Abstract

In the existing works for recommender systems, the design of Cross-Domain CTR prediction is less mentioned. In order to solve the data sparsity problem in CTR prediction, Cross-Domain Recommendation (CDR) leverages a wealth of information from a source domain to improve the CTR prediction performance on a target domain with sparse information, which is better than Single-Domain CTR prediction. Ads are usually displayed with natural content, which gives Cross-Domain CTR prediction the opportunity for problemsolving. In this paper, we propose a CDR novel named NAUI which can leverage auxiliary data information to improve the CTR prediction performance. NAUI utilizes an extracting user interest method which reduces excessive punishment for active users, and jointly training Cross Net and MLP components together on CDR. NAUI can capture more information about nonlinear features and combined features efficiently capture implicit and explicit high-order features interactions, and greatly improve the expression ability of the model on Cross-Domain CTR recommendation. Furthermore, we add an auxiliary classifier to the deep neural network to improve recommendation performance. Our experiment results has demonstrated that NAUI outperforms several frequently state-of-the-art methods of CTR prediction.