Research On Click-Through Rate Prediction Model Of Internet Advertising Based On Deep Learning

In the Internet era,the advertising industry is showing a trend of diversification and intelligence.With its low-cost operation model and good user experience,Internet advertising has become the most important form of advertising nowadays.In order to increase the number of user clicks and obtain traffic monetizing,the advertising system uses relevant strategies to predict the click probability of different ads,and selects ads with a higher probability value for targeted delivery.The accuracy of the click-through rate prediction can not only increase business revenue,but also increase user stickiness,so it has attracted wide attention from industry and academia.As the core of the advertising system,the click-through rate prediction model can deeply dig into the potential relationships between features and make accurate predictions of user click probability.From the perspective of feature combination,this paper addresses the shortcomings of existing methods,uses deep learning to build technical solutions,and conducts feasibility demonstration through experiments.The main work of the paper is as follows:(1)A Field-aware Bilinear Deep Factorization Machine(FBDFM)is proposed.The model introduces the bilinear perception layer to refine the feature crossing process and reduce information loss.At the same time,the model fully integrates the second-order combined features of multiple substructures,which reduces the model complexity of the neural network when performing high-order feature extraction and reduces unnecessary overhead.(2)An Attention-based Collaborative Factorization Machine(ACFM)is proposed.By improving the feature weighting module,the model distinguishes the effects of different features on the model effect,strengthens the ability to express important features,and makes the process of feature combination more interpretable.At the same time,the model fully integrates multi-level weight information for collaborative modeling through different attention substructures,which can effectively improve the prediction accuracy of the model.(3)A Deep Interest Perception Network(DIPN)is proposed.The model mines the implicit relationships between different behavior sequences,and obtains potential expressions of interest through the attention mechanism.Finally,it builds a prediction system together with the portrait features,which makes up for the shortcomings of the traditional feature modeling process that is difficult to effectively capture user preference.The paper has conducted research and experiments on the click-through rate prediction problem under sparse data.The above work is discussed in detail,and the feasibility of the model and the accuracy of the prediction are verified by the experimental results.