Deep Session Interest Network Based on the Time Interval Encoding for the Click-through Rate Prediction

Abstract

The click-through rate prediction is with significance in the field of recommendation systems, especially in advertising recommendation systems. At present, some sequence models based on deep learning have been directly used in the field of the click-through rate prediction to dig out the rule of user behavior and have achieved good results, but they ignored the influence of time information on the rule of user behavior. To solve the above problems, we propose a model named Time Interval Encoding Deep Session Interest Network (TIED-DSIN). In the TIED-DSIN model, a time interval encoding method is designed to integrate time interval information into the sequence model, and time decay factor is introduced in the encoding process to make the model consider the influence of time information fully when mining the rule of users' dynamic behaviors. Correspondingly, a comparative experiment is conducted on the real Alimama public data set, and the results show that the accuracy of the TIED-DSIN model is better than other models that commonly used in the click-through rate prediction.