Research On Advertising Click Rate Algorithm Model Based On Deep Learning

In online advertising applications for Internet marketing,ad click-through rate estimation can analyze user information,advertising merchants,search categories and contextual content in advertising commercial data,and accurate prediction results can reasonably adjust advertising strategies for Internet companies To improve users' online business experience.As deep learning has made great achievements in computer vision,natural language processing,and data mining,major companies,universities,and research institutions have also applied deep learning methods to ad click rate estimation,and have obtained certain effect.This paper summarizes the research on the cutting-edge deep learning-based advertising click-through rate estimation method and finds that the existing models are similar in structure.If you want to achieve better performance,you will innovate in the higher-order feature interaction part.The basic principle is to concatenate the feature vectors after Embedding,and then input them into a specially designed network to learn high-level feature interaction information.However,these methods have some shortcomings.The feature vectors after Embedding are only simple and unstructured,connected end-to-end,and the learning process is bit-by-bit and implicit when input to subsequent networks for high-order feature interaction.In general,this method cannot fully obtain high-level feature interaction information,and the interpretability is also weak.Based on this background,through the research and experimental exploration of various cutting-edge deep learning methods,this paper proposes a method of predicting the click rate using graph neural network.The combination of feature vectors is introduced into the graph structure.The feature vectors are used as The nodes in the graph use graph neural network to represent structured data very strongly to obtain higher-order feature interaction information more fully.In the graph neural network,each node represents the feature vector obtained after the original feature is subjected to Embedding,and the weight of the edge between the nodes indicates the importance between the features.Therefore,the high-order interactions between features are completed through the interaction of the edges of the nodes.Using graph neural networks,high-order feature interactions can be performed explicitly and efficiently,with higher interpretability.The whole work of this article includes the following three aspects:First,first introduce and summarize the current cutting-edge methods for predicting the click rate of ads based on machine learning and deep learning.Then,it introduces the commonly used methods of predicting the click rate of advertisements based on deep learning,graph neural network related technology and Attention related principles and technologies.Secondly,this paper applies the graph neural network GNN,which has outstanding performance in the field of computer vision,to the advertisement click-through rate prediction scenario.By using the Attention mechanism and LSTM technology,a new graph neural network structure AGNN is proposed.On this basis,combined with the classic ad click rate estimation method FM,a method of predicting ad click rate AGFM using graph neural network is proposed.Third,when conducting experiments,first compare the effects of setting different hyperparameters on the model's effect.Secondly,compare the experimental results with the model in this paper and some basic models.The model in this paper has improved to varying degrees in the evaluation indicators AUC and Logloss,so the effectiveness of the method proposed in this paper has also been verified.