Double-window advertisement delayed conversion prediction system

The invention provides a double-window advertisement delay conversion prediction system, which is characterized in that a preset reference delay display time is taken as a boundary to divide into data in a window and data outside the window, then a double-path neural network system is specially set, split training is carried out in a training process, and combined application is carried out in an application process. Therefore, the influence of the two kinds of data on the prediction result is fully considered, a sharing layer is introduced, the defect that common features of different data are lost due to independent training is overcome, and the system conversion prediction accuracy is further improved. 本发明提供一种双窗口广告延迟转化预测系统，以预设基准延迟显示时间为界限划分为窗口内数据和窗口外数据，然后为此特别设立了双路的神经网络系统，并在训练过程中进行拆分训练，在应用过程中组合应用，从而充分考虑两种数据对预测结果的影响，且在其中引入了共享层，弥补了单独训练所带来的不同数据共同特征丢失的缺陷，进一步提高了系统转化预测准确率。