Internet Advertiser Risk Forecasting Based On Cost - Sensitive Bayesian Classification

With the rapid development of social economy, and the rise of the Internet industry, online shopping has became a part of people’s daily life. And the online advertising market is already a 100 billion market. In this market, there are a lot of low-quality products, and advertisements. For all the websites, if all these low-quality advertisements display, they will face enormous legal risk. But the huge number of ads means there is no possible of human investigation. So for mathematical models, advertising risk prevention is very necessary. However, due to the time the Internet advertising market is very short. although many companies have the appropriate ways and means in its own operations, but there is little academic summary and demonstration, even can not find any articles in core journals This paper considered the initial research in this area. According to my 9 years of working experience in Alibaba’s advertising department, the risk of Internet advertisers is very similar with the financial corporate credit risk rating.In this paper, the experimental data comes from the sampling of a large online shopping website. In the progress of data preprocessing, this paper will compare some method of normalization and determined to user Feature scaling method to discrete data. A misjudgment must exist in classification problems, and different fault will cause different costs of loss. In the problem of online advertiser risk rating, a "normal" customer is mistaken for a "risk" customer and a "risk" customer is mistaken for a "normal" customer will come to different costs of loss. Although the former will lose some benefits, but the risk of loss the latter brought is immeasurable. This paper introduce the concept of cost sensitive into online advertiser risk rating problem, and construct the cost function and risk function. This paper will determine the optimal parameters of the cost function through lots of experiments, and establish the cost sensitive Naive Bayesian Model. The experimental results show that the cost sensitive Naive Bayesian Model can greatly reduce the probability of a III "risk" customer mistaken as a "normal" customer on condition that the correct rate was declined slightly.