Prediction and Application of Slot Attributes In Contract Display Advertisement

Abstract

In this paper, we propose a method that implicitly contains multiple features to describe the relationship between two variables, which was referred to as trend prediction. This method takes into account the relationship between the slot page view and multiple features. The data of different advertising slots are constructed by predicting methods such as tree model and neural network with category feature densification by auto-encoder model. To reduce the smooth time in the prediction curve, we take the logarithm function as a priori to regress the predicted curve. Finally, compared with our previous prediction model, the trend prediction model effectively improves the overall page view under the unconstrained mixed-integer programming model.