Research On Advertisement CTR Model Of Integrating Temporal Information

With the rapid development of the Internet,the online advertising model has gradually replaced the offline advertising distribution model.The task of online advertising is to predict the browser’s click probability on certain advertised items in a specific scenario,that is,the advertisement click-through rate estimation.In the task of advertising click-through rate estimation,the use of feature interaction to fully mine the implicit information between features is the main research direction,but such static construction of user and advertisement features cannot solve the problem of user interest drift.In response to this problem,this paper focuses on mining the time series information in the user behavior sequence while learning the effective feature information,dynamically constructing the user’s preference,and then obtaining a more accurate advertisement click-through rate estimation.The main findings of this paper are as follows:Firstly,according to the cyclic neural network that can process time series information,this paper uses a parallel structure to design an advertisement click rate prediction model that integrates time series information.The fusion of multi-characteristics in China.The advertising click-through rate prediction model integrating time series information mines low-level features and high-level feature information through the cross-network model,captures the changes in users’ long-term and short-term interests through the time series model,and integrates multiple feature information to improve the prediction of advertising click-through rate.Accuracy.Secondly,this paper introduces a gated neural unit with an attention mechanism on the basis of the above-mentioned advertisement click-through rate prediction model fused with time series information,and constructs an advertisement click-through rate prediction model fused with time series information of the attention mechanism.The model uses the gated neural unit with attention mechanism to obtain the contribution degree of user interest features in the long-term interest evolution process of users,that is,calculates the correlation between candidate advertisements and target advertisements,performs user interest screening,and recommends users to users.More interested in advertised items.Finally,in order to verify the effectiveness of the ad CTR prediction model fused with time series information and the ad CTR prediction model fused with attention mechanism time series information in the ad CTR task,the two models were used in two different advertisements.Experiments are carried out on the data set,that is,the ad click-through rate estimation model fused with time series information is compared with 4 common ad clickthrough rate estimation models,and the ad click-through rate estimation model with time series information fused with attention mechanism The model is compared with three ad click-through rate prediction models that introduce attention mechanism.