Design And Implementation Of Advertising Evaluation System

Today's "Internet mentality" is gaining traction,and most people are confused: how can so many free services generate revenue? If Internet APP is used as the front side of a variety of COINS,in fact,most COINS have the same pattern on the back side--a backward cash flow system with advertising as the core.Among them,free apps and services are the entrance of traffic and user data,while advertising business turns the incoming traffic and data into money,which is one of the most core models of the current Internet.This paper observes and analyzes the research status and application of mainstream advertising systems and advertising algorithms at home and abroad.This paper elaborates the demand analysis of advertising CTR estimation system,designs and implements the advertising CTR estimation system.The system has the functions of data collection,data preprocessing,macro statistical analysis of advertising data,CTR estimation of advertising plan,advertising audience estimation and so on.The system core module has(1)Advertising click through rate estimation module.This module USES deep spatial-temporal neural network to extract the feature input model of AD calendar click data,user data and AD data.The most widely used model in the advertising business only studies a single target AD,ignoring the impact of other ads related to the target AD on the click-through rate of the target AD.More other advertising data are added to the deep spatial-temporal neural network DTSN to assist in the prediction of the click-through rate.This model USES Avito advertising data set,search advertising data set,and information flow advertising data set to conduct model training and effect verification.In the training process,neural network of different layers and dropout ratio are used for effect tuning.Experiments on the above data set show that the DSTN model is superior to the most advanced models in CTR estimation.(2)Advertising audience estimation module.In the module used in the automatic encoder model,figure convolution is based on the principle of main network and two models,the automatic encoder figure convolution networks with the advantages of high efficiency and high accuracy,advertising the interaction with the user data can easily be converted into graph structure,in figure structure on machine learning is a very difficult task,because of the complexity of the structure is higher,the information is rich,while using convolution networks can complete study of the structure of figure,can directly process diagram and use the structure information.The model USES Movie Lens,Douban,and information stream advertising data sets for model training and effect verification.