Click rate estimation method based on deep learning and information fusion

Abstract

The invention provides a click rate estimation method based on deep learning and information fusion, and the invention is characterized in that the invention is divided into three modules: a field domain decomposition machine FFM of a shallow extraction module, a convolutional neural network CNN of a deep extraction module, and a deep belief network DBN of a feature fusion module, wherein the shallow layer module and the deep layer module adopt a parallel structure and share a fixed dense vector converted from discrete features of users and commodities; the shallow layer module adopts a second-order combination of FFM automatic feature extraction, the deep layer module adopts a CNN local sensing domain to extract a high-order nonlinear feature combination, and the fusion module adopts DBNto fuse the output of the shallow layer FFM and the deep layer CNN to realize the interaction between the shallow layer features and the deep layer features. According to the invention, the internal relation between the characteristics is mined in combination with the characteristic interaction depth of the shallow layer and the deep layer, the problems of gradient explosion and gradient disappearance are effectively solved, and the click estimation capability is improved.

Description

technical field [0001] The invention relates to a method for estimating click rate, in particular to a method for estimating click rate based on deep learning and information fusion, and belongs to the field of recommendation systems. Background technique [0002] With the combination of deep learning and recommendation system, the method of click rate estimation has also undergone earth-shaking changes. From the initial combination of logistic regression and gradient boosting tree combination of artificial features to the factorization machine FM of shallow feature automatic combination, Huawei Noah's Ark proposed deep learning as deep Deep FM, and the accuracy of click rate prediction has been significantly improved. However, there are still some problems in the existing click-through rate estimation methods. The deep DNN network will have the problem of gradient explosion and gradient disappearance as the number of layers increases, resulting in difficulties in training a...

AI Technical Summary

Problems solved by technology

From the initial combination of logistic regression and gradient boosting tree combination of artificial features to the factorization machine FM of shallow feature automatic combination, Huawei Noah's Ark proposed deep learning as deep Deep FM, and the accuracy of click rate prediction has been significantly improved. However, there are still some problems in the existing click-through rate estimation methods. The deep DNN network will have the problem of gradient explosion and gradient disappearance as the number of layers increases, resulting in difficulties in training and optimization.

Images