Research On Online Advertising CTR Prediction Based On Improved Time Series Algorithm

This time,the rapid development of the Internet has greatly changed people's lives,and online advertising as the largest source of Internet profits has attracted much attention.Both the academic and industrial sectors have invested a lot of manpower and material resources for the development of advertising technology.Under the impact of new technologies based on machine learning,online advertising push and user data mining have entered a new era of artificial intelligence.Online advertising is an emerging research topic involving,machine learning,Natural language processing,psychology,and even micro-economics and many other fields of knowledge.Traditional advertising CTR estimation is based on manual rules and basic time slicing methods to directly model advertising data,which brings a series of problems,such as slow model running,sparse advertising features,cold start,low-achieving ad CTR estimation and so on.To solve the above problems,this paper uses the methods of machine learning and data mining to extract data features from massive data based on the improved time slice algorithm,and refers to a factor decomposition method for the extracted new features.The collaborative filtering problem of traditional advertisement clicks is transformed into the user-article pairing problem,which greatly reduces the sparsity of high dimensional features.In addition,this paper also proposes some optimized feature extraction methods based on time-slicing.At last,XGBoost,GBM and MLP are applied to verify the data features.The experimental data source is Jing Dong JData algorithm contest real user data.The experimental results show that the improved time-slicing algorithm presented in this paper has greater progress in the click-through rate pre-problems than the traditional methods,and it has excellent performance in JD.com official evaluation indicators,AUC values,ROC curves and other evaluation indicators.