CTR Prediction with User Behavior: An Augmented Method of Deep Factorization Machines

In Click-Through-Rate(CTR) prediction problems, video and comment thumbs-up prediction has important commercial values. The click rate prediction system can find the target population more accurately, which not only improves the user experience but also brings profits to the companys. Click behavior has a great correlation with personal habits. The Deep Factorization Machine(DeepFM) model is at the cutting edge of CRT by learning both high-level and low-level features. But it cannot hightlight the impact of user's personal behavior with has a great correlation with click behavior. In this paper, we propose behavior-empahsized algorithms, DeepFM & Habit (Pro) and DeepFM & Habit (LR), for different situations. The experimental results on TikTok dataset (short video data) demonstrate that our proposed methods achieve superior performance with less time consumption on this CTR prediction problems.