O2O场景下的反作弊分析模型的设计与实现

发布时间：2018-06-29 22:01

  本文选题：反作弊 + 数据挖掘　； 参考：《哈尔滨工业大学》2016年硕士论文

【摘要】：随着Internet和相关Web技术的发展,新的电子商务交易模式悄然兴起。近年来,Online To Offline(O2O)模式飞速发展。O2O模式是一种将线下交易与互联网结合在一起的新的商务模式,即线上网站通过提供打折、返利补贴、提供送货服务等方式,把线下商店的消息推送给线上用户,用户在选定相关商户之后在线下单、在线支付等流程,之后再凭借订单去线下商家提取商品,或等待送货上门的服务,或享受线下其他服务[1]。O2O市场中,各大公司为了抢占市场份额,纷纷推出了各种补贴机制来吸引用户使用本公司产品。而对应的,“刷单”这一行业也是O2O公司所必须面对的问题。而目前刷单主要靠人工线下检查为主,如监控某个地区的单量异常,某家店铺的消费均量异常等,然后进行线下检查。作弊成本低,监察成本高,是现在O2O公司所面临的最大的问题。本课题基于某餐饮外卖O2O公司的反作弊部门,从打击刷单的需求点出发,提出了一种可以通过机器学习和数据挖掘相结合的方法,来检测某一用户为刷单用户的风险,从而降低监察的成本。本课题首先论证了O2O的作弊现象和网页排名作弊现象的异同,并针对网页排名的反作弊方法进行了修改使其契合本课题所面对的问题。同时,由于要对高风险用户进行进一步操作,本文还开发了相应的后台操作模块,配合其他提高作弊成本的方式构成监察系统,从而降低刷单比例。数据平台使用了敏捷开发的策略,同时使用了分布式数据库等技术实现了自由维度组合生成报表的需求。目前,反作弊系统已经上线五个月,并历经两个版本的升级,数据表明,本文的方法可以有效的识别高风险用户,并且提升了线下监察部门的工作效率,有效降低了监察成本。
[Abstract]:With the development of Internet and related Web technology, the new electronic commerce transaction mode rises quietly. In recent years, the online to offline (O2O) model has developed rapidly. The O2O model is a new business model that combines offline trading with the Internet, in which online websites offer discounts, rebate subsidies, delivery services, etc. Push the message from the offline store to the online user. After selecting the relevant merchant, the user sends out an order online, pays online, etc., and then relies on the order to pick up the goods from the offline merchant or to wait for the service to be delivered to the door-to-door. Or enjoy other services offline [1] .O2O market, in order to seize market share, companies have introduced a variety of subsidy mechanisms to attract users to use their products. And corresponding, "brush order" this industry also is the problem that O 2 O company must face. At present, the brushing order mainly depends on the manual line inspection, such as monitoring the single quantity anomaly in a certain area, the average consumption quantity of a shop, etc., and then carries on the offline inspection. Low cost of cheating and high cost of supervision are the biggest problems faced by O 2 O companies. Based on the anti-cheating department of a restaurant takeout O2O company, this paper presents a method that can be combined with machine learning and data mining to detect the risk of a certain user as a brushing user from the point of view of attacking the demand for brushing. This reduces the cost of monitoring. This paper first demonstrates the similarities and differences between the cheating phenomenon of O2O and the cheating phenomenon of the web page ranking, and modifies the anti-cheating method of the web page ranking so as to fit the problems faced by this subject. At the same time due to the high risk users to further operation this paper also developed the corresponding backstage operation module with other ways to increase the cost of cheating to form a monitoring system so as to reduce the proportion of brush orders. The data platform uses the strategy of agile development and the technology of distributed database to realize the requirement of generating report by free dimension combination. At present, the anti-cheating system has been on line for five months, and after two versions of the upgrade, data show that the method can effectively identify high-risk users, and improve the efficiency of offline supervision departments, effectively reduce the cost of supervision.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.52
，

本文编号：2083515