Multi-target sorting model training method and device and user behavior prediction method and device

The invention provides a multi-target sorting model training method and device and a user behavior prediction method and device, and relates to the field of artificial intelligence. The method comprises the following steps: constructing an initial multi-target sorting model and a loss function; firstly, processing a training sample through the initial multi-target sorting model to obtain the clickestimation value and the multi-task estimation value, and then correcting the click estimation value, so that the deviation between the click estimation value and the actual click rate is reduced. The problems of inaccurate prediction and large prediction error when the multi-target sorting model obtained through training predicts the user behavior in actual application can be avoided. 本申请提供一种多目标排序模型训练、用户行为预测方法及装置，涉及人工智能领域。在构建初始多目标排序模型和损失函数之后，先通过初始多目标排序模型对训练样本进行处理，得到点击预估值和多任务预估值，然后再对点击预估值进行矫正处理，以减少点击预估值与实际点击率之间的偏差，能够避免训练得到的多目标排序模型在实际应用中预测用户行为时，出现预测不准确，预测误差大的问题。