Research On Advertising Exposure Prediction Model Based On Improved Deep Learning

With the rapid development of Internet technology,online advertising system is an important source of income for various Internet companies.Advertising exposure prediction is a relatively important part of the online advertising system.It can provide advertisers with a reference for the future advertising exposure performance.Through this prediction,advertisers can avoid blind optimization attempts and reduce the cost of trial-and-error.The difficulty in computational advertising lies in the large amount of data,the high dimension of data,and the high correlation between different features.Therefore,how to deal with high-dimensional data quickly and do feature interaction automatically is the key to solve the problem.Based on deep learning models commonly used in the field of computational advertising,a new neural network structure is proposed.First,due to the high dimensional and sparse features and high interaction between features,the fusion factorization machine and attention mechanism module are introduced to effectively obtain the low degree interaction relationship of the high dimensional and sparse features.The user's exposure record is used as a corpus for word embedding using Word2 Vec method,and the vector obtained after training is directly input to the deep network part.At the same time,batch normalization is used in the deep network part to increase the number of network layers and the degree of feature interaction.There are three main advantages of the improved model.First,after considering that different combinations of features have different importance,and introduces attention mechanism to assign different weights to different combinations of features,which improves the expression ability of the model.Second,batch normalization is used to deepen the deep network part and introduce high degree feature interaction.Third,using Word2 Vec to get the embedding improves the model's computing speed and effect compared to using it directly.Finally,the data of 2019 Tencent advertising algorithm competition is used to test the improved model and other models.Experimental results show that the improved model reduces the mean square error index by 3.7% compared to the Light GBM model,and the computing speed is increased by 54.1%.The experiment verified the effectiveness and feasibility of the improved model.In order to further improve the prediction accuracy of the model,an advertising exposure prediction method based on model fusion is adopted.The experimental results show that the mean square error of model fusion is 0.7% lower than that of the improved model.