Online Advertising Click Fraud Monitoring And Countermeasure Research

With the growth of China's economy,the proportion of domestic online advertising service expenses has also been increased in recent years.Advertisers publish online advertisements through advertising service providers in the online advertising media(various APP ? websites,etc.).After the successful release,advertisers pay for the user's click behavior,advertising media to obtain advertising fees.Some advertising medias often use various means to create click fraud in order to obtain more clicks.According to the relevant report,the proportion of abnormal traffic in online advertising reached 30.2% in the whole year of 2018,nearly half of which came from click fraud,which shows that the phenomenon of click fraud in online advertising is more serious.Therefore,for the network advertising click fraud monitoring governance is urgent.Based on the research of click fraud literature at home and abroad,this paper aims to solve the problem of advertisers' monitoring and governance of online advertising.Aiming at the pain point of advertisers,this paper puts forward a question: How to find a suitable technique to identify click fraud accurately and cheaply,and how to improve the situation of click fraud by implementing appropriate countermeasures.At the same time,in order to solve this problem,this paper also refines three key factors: 1 click fraud identification technology factor 2 click fraud identification cost factor 3 click fraud governance factor.Aiming at the optimization of the first factor,the fourth chapter innovatively uses the convolution denoising self-encoder model.Network advertising data sets,the normal sample generally has the same characteristics,and abnormal or different causes of fraud,showing different characteristics of fraud.Due to the low proportion of abnormal data,the BuzzCity click fraud database cited in this paper can only learn the effective features from the normal sample data,and compute the reconstruction error.Finally,the simulation results show that the proposed method achieves more than 90% in all indexes,and the recall rate exceeds nearly 10% of the other three ways.This can help advertisers preliminary screening out which is normal click,which is an abnormal click.In view of the second factor,this paper puts forward an innovative end-to-end click fraud monitoring method based on Semi-supervised GAN(SSGAN).Compared with the traditional semi-supervised learning method,the method based on SSGAN can only use shallow features of data,and can fully extract the internal relationship of data by mining the deep features of data,realize click fraud monitoring based on some existing class mark data and a large number of non-class mark data,enhance the generalization ability of the model,and greatly improve the efficiency.In order to reduce the proportion of expert manual labeling and keep high detection accuracy,the cost of labeling is greatly reduced.Because domestic fraud is more severe than in Singapore,experts have labeled more online advertisers with fraudulent features,so there are enough anomalous clicks to learn.So we can learn the characteristics of normal click and abnormal click respectively to identify click fraud.Therefore,the fifth chapter further expands and optimizes the first influencing factor of click fraud identification technology.Aiming at the optimization of the third factor,this paper reduces the proportion of abnormal flow of click fraud from the source by adding measurement index,contract,network advertising media evaluation process and cooperation with network advertising service providers.Through the use of innovative algorithms and efficient means of governance,this paper solves the biggest problem that puzzles advertisers: how to find the right technology and low-cost to identify click fraud,and improve the situation of click fraud through the implementation of appropriate governance measures.