Multi-Task Deep Learning with Task Attention for Post-Click Conversion Rate Prediction

Online advertising has gained much attention on various platforms as a hugely lucrative market. In promoting content and advertisements in real life, the acquisition of user target actions is usually multi-step process, such impression→click→conversion, which means process from delivery recommended item to user’s click final conversion. Due data sparsity or sample selection bias, it difficult for trained model achieve business goal campaign. Multi-task learning, classical solution this problem, aims generalize better original task given several related tasks by exploiting knowledge between share same feature label space. Adaptively learned relations bring performance make full use correlation tasks. We train general capable capturing relationships all existing active meta-learning perspective. addition, paper proposes Attention Network (MAN) identify commonalities differences The improved explicitly learning stacking To illustrate effectiveness our method, experiments are conducted Alibaba Click Conversion Prediction (Ali-CCP) dataset. Experimental results show that method outperforms state-of-the-art multi-task methods.