Click rate estimation model based on feature representation under attention mechanism

Abstract

In order to complete click rate estimation according to object features of an object to be detected, the invention can be applied to the fields of enterprise-level recommendation systems, search systems, online advertisement systems and the like as a data fine arrangement link. The invention provides a click rate estimation model based on feature representation under an attention mechanism. The model comprises a feature embedding layer, and the feature embedding layer is used for carrying out the vectorization processing of continuous features and discrete features so as to form a stacked feature and explicit feature cross network, performing explicit feature combination and an implicit feature cross network on the stacking features through the attention cross network, performing implicit feature combination on the stacking features through the multi-layer perceptron, estimating a probability output layer, and estimating the click rate according to the received combined features, wherein the attention crossover network eliminates the dependence of an estimation model on artificial feature engineering, and meanwhile, due to the introduction of an attention mechanism, the importance of each combination feature on model estimation is distinguished, and the influence of useless and redundant features on the model is eliminated.

Description

technical field [0001] The invention belongs to the technical field of data mining, and specifically relates to an end-to-end click prediction technology, which uses a deep learning model to automatically complete feature representation and then predict the click rate model. Background technique [0002] As a key technology that directly affects user platform experience and advertising revenue, click-through rate estimation has always been one of the core research topics in the industry. At present, research work at home and abroad is mainly at the level of feature representation. Existing methods are mainly divided into machine learning click rate models and deep learning click rate models. [0003] In the early stage, the industry was limited by computing power, online learning, and model deployment, mainly by building lightweight machine learning models. The most classic model is the Logistic Regression model (Logistic Regression). LR has quickly become the mainstream mo...

AI Technical Summary

Problems solved by technology

In 2010, Brendan McMahan et al. proposed the online learning algorithm FTRL (Follow The Regularized Leader) for LR, which further promoted the application of LR in the industry. However, the LR model is essentially a linear model with limited learning ability, and its prediction effect usually depends on data scientists. feature engineering capabilities

[0007] In summary, the current deep learnin

Images