Research On Click-Through Rate Prediction Models Based On Deep Neural Networks

Click-through rate prediction has always been a hot issue in the field of online advertising,which can improve the advertiser's revenue and provide users with a better experience.Although some scholars have applied the factorization machine to the click-through rate prediction model and proposed a neural network-based method,the prediction accuracy of the model on high-dimensional and high-sparse advertisement data is still not high.Therefore,we have started a series of researches on these issues.The following are our main research work and innovations:1.At first,this paper introduces the relevant knowledge of CTR prediction and the research background in this direction,and analyzes the current research status at home and abroad.Secondly,it introduces a series of technologies related to click-through rate prediction.The most important is to summarize the click-through rate prediction method based on factorization machine and explain the commonly used evaluation indicators in CTR prediction.Finally,this paper details the click-through rate prediction based on support vector machine.2.Because deep neural network exhibits powerful function in mining feature interaction,so the paper introduces the click-through rate prediction model that based on deep neural network.The model can well mine high-order feature interactions and low-order feature interactions,while sharing the same input without any feature engineering.This model improves the accuracy of click-through rate prediction to some extent.3.At present,most CTR prediction models only consider how to mine feature interactions,and ignore the impact of user interest on the accuracy of CTR prediction.Therefore,this paper proposes a click-through rate prediction model based on LSTM.The model uses the long short-term memory to extract user interest features and combines other features for click-through rate prediction.Compared with other existing models,the proposed model improves the accuracy of CTR prediction.4.Based on this,combining the attention mechanism to capture the evolution of user interest and proposes a novel attentive deep interest-based network model called ADIN.First,wecapture the interest sequence in the interest extractor layer,and the auxiliary losses are employed to produce the interest state with the deep supervision.Next,we extract the interest evolving process that is related to the target and propose an interest evolving layer.At the same time,attention mechanism is embedded into the sequential structure.Then,the model learns highly non-linear interactions of features based on stack autoencoders.Experimental results show that the model improves the accuracy of click-through rate prediction.In summary,based on the problems existing in the process of CTR prediction,this paper proposes a model of CTR prediction.This model is mainly based on deep neural networks,which can better mine hidden feature interactions.Finally,the effectiveness of the research work is proved through theory and experiments.The proposed model improves the accuracy of CTR prediction to a certain extent.