A Prototype Data Management Platform with Machine Learning-based Automated Turing Test for Prevention of Online Advertising Fraud

具有基于机器学习的自动图灵测试的原型数据管理平台，用于预防在线广告欺诈

• 批准号: 720789
• 金额: $ 31.83万
• 依托单位: MEDIAGAMMA LIMITED
• 依托单位国家: 英国
• 项目类别: GRD Development of Prototype
• 财政年份: 2016
• 资助国家: 英国
• 起止时间: 2016 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=720789
• 关键词: Prototype; Data; Management; Platform; Machine

Over 29% of internet activity arises from malicious non-human traffic (NHT) andprogrammatic platforms for automated placement of online display advertising via real-timebidding (RTB) are highly vulnerable to Click Fraud. Click Fraud occurs when bots falselygenerate impressions or enhance the click-through rate of displayed ads, generating fraudulentpay-per-click or cost per mille revenues by artificially increasing the bid value/frequency.Advertising fraud is hard to detect and leads directly to >$68bn/yr of lost revenues byadvertisers/media agencies, as well as restricting the £24bn/yr opportunity to leverage valuefrom analytics data for personalised marketing.Existing software to prevent CF uses rules-based filtering, restricted for RTB by limitedcapability to detect the origin of a click/impression (ground truth) for labelling NHT (negativeexamples) only after a visit/bid. Limited platform compatibility and rule generation rates arealso unable to address a rapidly evolving Click Fraud threat.In contrast, MediaGamma Limited (MG) aim to develop a prototype Automated Turing Testas a pre-bid solution to prevent Click Fraud using 1 million binary bid request features, byactively and continuously identifying the characteristics of fraudulent clicks/impressions fromNHT. Using machine learning algorithms integrated through a novel Data ManagementPlatform for RTB ad placement, trained via techniques inspired from biological behaviouraltests, MediaGamma target capability to prevent an agency/advertiser from participating in afalse bidding process in real-time. This 12 month project will cost £555,152 withperformance of the prototype Data Management Platform software module demonstratedthrough A/B testing in commercially relevant situations.

超过29%的互联网活动来自恶意非人类流量(NHT)，通过实时竞价(RTB)自动投放在线展示广告的编程平台非常容易受到点击欺诈的攻击。当机器人虚假地产生印象或提高显示广告的点击率时，就会发生点击欺诈，通过人为地提高竞价值/频率来产生欺诈性的按点击付费或每百万收入成本。广告欺诈很难被发现，直接导致广告商/媒体代理每年损失680亿美元的收入，并限制了每年240亿GB的机会利用分析数据的价值进行个性化营销。现有的防止点击欺诈的软件使用基于规则的过滤，限制RTB只有在访问/竞标后才能检测点击/印象(基本事实)的来源来标记NHT(否定的例子)。有限的平台兼容性和规则生成率也无法应对快速发展的点击欺诈威胁。相比之下，MediaGamma Limited(MG)的目标是开发一个原型自动化图灵测试，作为投标前的解决方案，通过积极和持续地识别来自NHT的欺诈性点击/印象的特征，使用100万个二进制投标请求特征来防止点击欺诈。MediaGamma使用通过用于RTB广告投放的新型数据管理平台集成的机器学习算法，通过生物行为测试启发的技术进行培训，以确定目标能力，以防止代理/广告商实时参与公平的竞价过程。这一12个月的项目将耗资555,152 GB，原型数据管理平台软件模块的性能通过A/B测试在商业相关情况下得到证明。