Design And Implementation Of Marketing Activities Risk Control System

In order to sell and promote a product, companies usually do some marketing activities for the product. This initiative is not only to let consumers understand the product deeply, but also to explore their deeper level of consumer demand. Tencent company gets that as well. She usually promote her products through giving users Q coins, game currency, virtual props, etc. However, in the face of these high value marketing resources, a career called "wool party" comes out. They selectively participate in the company’s online activities, obtain the marketing resource with relatively low price or even zero cost, then they sell the resource immediately and get rewarded financially. They broke the rule seriously, and plundered the activities resources, so that companies have to increase the cost of products continuously. They seriously damaged the reputation and image of the company, and make the company face the huge financial risk. In view of this phenomenon, we have developed a risk control system for marketing activities, to prevent these criminals through technical means.First problem to be solved is to identify these criminals, in technical point of view is to identify abnormal account. We designed and implemented the “abnormal account modelling” subsystem. Based on some abnormal accounts information, using C4.5 decision tree algorithm of machine learning, we have got the model of the abnormal accounts’ characteristics. Then we designed and implemented “historical risk account analysis” subsystem based on Hadoop. After users participated in marketing activities, we analyzed all the user who participated in the activities, and found out the accounts which had malicious behavior in the active, so we can close their accounts or take out their game coins. We also designed and implemented the “real-time interception” subsystem based on Kafka messaging framework. Before user participated in marketing activities, we analysised the user’s transaction flow, stop the risk ongoing transaction timely, protect property and interests of the company more effectively. According to the evaluation indexes of machine learning field, we evaluated the model and obtained good results. In fact, after system running on-line, we rarely received complaints from users, so we could see that our system is very effective in another view.Through our risk control system, the threshold of criminals getting profit is getting higher and higher, and benefits are also fewer and fewer. So we have cracked down on this behavior effectively, and saved a huge loss for the company.