Research And Application Of Intelligent Recommendation System Based On User Behavior Characteristics

After decades of amazing development,the Internet,cloud computing and AI have covered every aspect of our daily life.Intelligent recommendation technology,which has great commercial potential,is valued by major Internet companies,and there are numerous researches related to it.Under such advantaged conditions,intelligent recommendation system can easily analyze each person’s preferences and status,and give the most appropriate opinions for people’s choices.In the research of intelligent recommendation system,it has gone through several stages of development.From the early use of collaborative filtering model for recommendation,to the rise of machine learning model to try to integrate machine learning,and finally CTR（Click through rate）recommendation model stood out and surpassed the classical machine learning model.This thesis focuses on the CTR model,introduces high-order dominant feature cross term into the CTR model,and makes improvement research on the model and applies it to the commodity recommendation system.The main research contents of this thesis include as follows:1.The CTR recommendation model of the current classic Wide and Deep architecture is analyzed.In the face of the conflict between the pursuit of high universality and fast data fitting,this kind of architecture model can well grasp the balance between the two to achieve better performance.On this basis,CIN（Compressed interaction network）network structure is analyzed and introduced.CIN model can effectively utilize the high-order dominant interaction between input features,which not only greatly enhances the interpretability of the model,but also greatly improves the accuracy of the model.2.For CTR model introduced into CIN network,experimental research was conducted on criteo advertising data set.Firstly,the hyperparameters such as CIN network layers,feature map dimension and regularization coefficient of the model were optimized to explore the potential performance of the model.Then,the input and output structures of the model are optimized to make the model more effective in processing complex and diverse data inputs and make full use of the processing results of CIN model.Finally,through the comparison of experimental results,it is proved that the improved CTR model has been significantly improved in logloss and AUC indexes compared with the model without improvement.3.Based on the above research and combined with practice,a financial information recommendation system based on CTR model is established to analyze user behavior characteristics.The recommendation system can provide reasonable information recommendation by combining the basic information of users with the behavioral characteristics of user groups such as click,purchase and collection.This system mainly focuses on solving the application problems such as how to collect and store CTR data set in web system,how to use CTR model to realize recommendation function,and how to use artificial feature training model reasonably...