基于用户相似度和特征分化的广告点击率预测研究

大数据环境下如何对互联网广告进行精准投放一直是计算广告学领域高度关注的问题。作为在线广告投放效果的一个重要指标,点击率的精确预测关系到媒体、用户和广告主三方的利益。目前的主流方法是通过抽取特征建立单一点击率预测模型,其不足之处在于使用单个权重来度量特征对点击率的影响过于片面。该研究基于分而治之的思想,提出了基于用户相似度和特征分化的混成模型。该模型首先根据混合高斯分布来评估用户相似度,将其划分为多个群体。针对不同群体,分别构建子模型并进行有效组合,从而挖掘同一特征对不同群体的差异化影响,进而准确地预测广告点击行为。通过使用真实互联网公司的广告数据集进行实验,并与主流方法做了详细的对比分析,检验了该方法的有效性。     

AdROSA—Adaptive personalization of web advertising

One of the greatest and most recent challenges for online advertising is the use of adaptive personalization at the same time that the Internet continues to grow as a global market. Most existing solutions to online advertising placement are based on demographic targeting or on information gained directly from the user. The AdROSA system for automatic web banner personalization, which integrates web usage and content mining techniques to reduce user input and to respect users’ privacy, is presented in the paper. Furthermore, certain advertising policies, important factors for both publishers and advertisers, are taken into consideration. The integration of all the relevant information is accomplished in one vector space to enable online and fully personalized advertising.     

基于用户行为分析的广告欺诈点击检测

在网络广告业中出现的欺诈点击行为,使得搜索引擎企业以及广告主的利益受到了严重损害,致使点击付费模式遭到质疑,欺诈点击已经成为阻碍网络广告业健康发展的一大顽疾。针对网络广告业发展所面临的此种困境,提出一种基于用户行为分析的广告欺诈点击检测技术。首先创建用户行为数据仓库,然后运用贝叶斯分类方法对用户行为数据进行点击合法等级预测,最后结合博弈控制机制对用户点击有效性进行最终判断。     

点击欺诈群体检测与发现*

在线广告中的欺诈点击(click fraud)是指所有利用欺诈性手段或带有欺诈意图并被搜索引擎承认的点击行为。传统点击欺诈检测主要集中在检测个体用户点击的合法性。然而,目前存在很多的发布商雇佣大批网络用户,以群体形式进行欺诈点击。针对这一问题,提出了一种检测点击欺诈群组的方法。首先使用频繁项集挖掘算法来发现共同点击过大量广告的个体用户,作为疑似欺诈组。然后,在对组内用户点击行为属性分析的基础上,运用孤立点检测方法找到与组内其它用户有显著差异的疑似欺诈用户。最后,运用贝叶斯分类方法对检测到的所有疑似欺诈成员分类,得到真正的欺诈群组和欺诈用户。在真实数据集上的实验结果证明了方法的可行性与有效性。     

基于模糊优化的关键词广告预算分配策略

针对单搜索引擎,研究了广告主竞投多个关键词时的广告时序预算分配策略,在总预算限制下建立了以最大化广告收益、最小化无效点击为双目标的单引擎多关键词广告时序预算分配模型,并给出了模糊优化双目标预算分配模型的解法。通过验证,该预算分配模型对广告主在单搜索引擎上竞投多个关键词广告时起着策略性指导作用。     

A software infrastructure for authenticated Web metering

As the popularity of the Internet and the number of resources available on it have grown, potential customers are increasingly turning to it for information about products and services. Accordingly, online advertising is gaining a significant portion of the advertising market. The Internet has become a mainstream advertising channel, surpassing traditional media such as newspapers and radio in number of advertisements. Advertisers exploit the popularity of the best-known Web sites, typically search engines or portals, to advertise their products and reach the most potential customers. Traditional rating systems are of little value when applied to the Internet because of the enormous number of Web pages available to online advertisements. Counting accesses to a Web service is a difficult task and the data may be unreliable. Several metering techniques attempt to accurately measure the number of visits a site receives and hence the advertising exposure, but advertisers and auditing companies haven't adopted a standard technique. We propose a framework based on hash chains. Unlike similar approaches, our implementation minimizes the overhead associated with the additional communication required to implement the protocol while providing an efficient and flexible scheme. Furthermore, the resulting framework offers additional guarantees such as security and nonrepudiation of the produced proof of visits.     

An analysis of the importance of the long tail in search engine marketing

Search engine marketing is currently the most popular form of online advertising. Many advertising agencies and bloggers claim that the success of search engine marketing is driven by the “long tail”, defined in this research as the many less popular keywords employed by users to search the Internet, and suggest that search engine marketing campaigns need to have hundreds or thousands of keywords to accommodate this phenomenon. We doubt this claim. Our data from three search engine marketing campaigns in two countries, which report the success of a total of 4908 keywords over 36 weeks, covering 10,104,015 searches and 492,735 clicks, show that the top 20% of all keywords attract on average 98.16% of all searches and generate 97.21% of all clicks. The use of the top 100 keywords in each campaign attracts on average 88.57% of all searches and 81.40% of all clicks. These results are fairly stable across a varying total number of keywords in use and suggest that the success of search engine marketing is driven by relatively few keywords. However, we also show that the set of the top 100 keywords changes over time. Hence, advertisers need to concentrate on finding the top 100 keywords, but they do not need to bother too much about the performance of keywords in the long tail.     

The key role of relevance in personalized advertisement: Examining its impact on perceptions of privacy invasion,self-awareness,and continuous use intentions

Internet advertising has become increasingly personalized as advertisers tailor content to individual users. However, this has led consumers to be concerned about their privacy. Based on rational choice theory and self-awareness theory, the current research explores the role of relevance in personalized advertisements and examines its impact on perceptions of privacy invasion, self-awareness, and subsequent continuous use intentions of personalized advertising. Analysis of survey data from 386 online users found that although privacy invasion perceptions are negatively related to continuous use intentions, perceived advertisement relevance mitigates consumer's privacy concerns. Perceived relevance was also found to be positively related to consumer's continuous use intentions through the mediation of self-awareness. This research identifies and highlights the importance of relevance in the tension between privacy concerns and personalized advertisements.     

Optimal Scheduling and Placement of Internet Banner Advertisements

The increasing popularity of the World Wide Web has made it an attractive medium for advertisers. As more advertisers place Internet advertisements (hereafter also called "ads"), it has become important for Web site owners to maximize revenue through the optimal selection and placement of these ads. Unlike most previous research, we consider a hybrid pricing model, where the price advertisers pay is a function of 1) the number of exposures of the ad and 2) the number of clicks on the ad. The problem is finding an ad schedule to maximize the Web site revenue under a hybrid pricing model. We formulate two versions of the problem - static and dynamic - and propose a variety of efficient solution techniques that provide near-optimal solutions. In the dynamic version, the schedule of ads is changed based on individual user click behavior. We show by using a theoretical proof under special circumstances and an experimental demonstration under general conditions that a schedule that adapts to the user click behavior consistently outperforms one that does not. We also demonstrate that to benefit from observing the user click behavior, the associated probability parameter need not be estimated accurately. For both of these versions, we examine the sensitivity of the revenue with respect to the model parameters.     

Web挖掘在网络广告点击欺诈检测中的应用

为有效解决网络广告中存在的点击欺诈问题,提出了一种基于Web挖掘算法的解决方案,并设计了一套点击欺诈检测模型.该模型通过对点击流进行时序分析、离群点挖掘、非线性分析等操作,能有效检测或屏蔽各类点击欺诈,有效屏蔽无意识的无效点击,并且在不影响广告展示速度的基础上显著提高检测点击欺诈的效率.实验结果表明,该解决方案可以有效检测采用手动或者利用计算机程序的方法模仿正常用户进行点击欺诈的行为,表明了模型的可行性和方案的有效性.     

An empirical test to forecast the sales rank of a keyword advertisement using a hierarchical Bayes model

Online advertising (ad) is a form of promotion that uses the Internet and World Wide Web for the expressed purpose of delivering marketing messages to attract customers. Not surprisingly, how to predict the effectiveness of online advertising has gained lots of research attention. This study introduces the hierarchical Bayesian analysis to the online advertising effect model involving competition with other products. It developed a competition model with a time-decaying effect that is applicable for the sales-rank data in the online marketplace. The proposed model formalizing the hierarchical structure has performed better than the reduced model without having random effect components. It captures the heterogeneous advertising responses across the products as well as search keywords. Our results have implications for online advertising effect measurement, and may help guide advertisers in decision-making.     

一种有效预防点击欺诈的算法

网络广告中,点击欺诈愈演愈烈,已经成为阻碍网络广告业健康发展的一大顽疾。针对网络广告业发展所面临的困境,对预防点击欺诈进行了研究,提出一种基于浏览时间和点击频率,并与验证码相结合的算法。该算法能有效屏蔽类似于木马点击器软件的点击欺诈,有效屏蔽浏览者偶然的无意识的无效点击,显著降低人工点击欺诈的效率,同时也不会让真正的潜在客户流失。     

一种有效预防点击欺诈的策略

网络广告中,点击欺诈愈演愈烈,使得搜索引擎企业以及点击付费模式遭到质疑。点击欺诈已经成为阻碍网络广告业健康发展的一大顽疾。针对网络广告业发展所面临的困境,提出一种基于图形验证码的预防点击欺诈策略。该策略能有效屏蔽类似于木马点击器软件的欺诈点击,有效屏蔽浏览者偶然的无意识的无效点击,显著降低人工欺诈点击的效率。     

CAT-RFE:点击欺诈的集成检测框架

点击欺诈是近年来最常见的网络犯罪手段之一,互联网广告行业每年都会因点击欺诈而遭受巨大损失。为了能够在海量点击中有效地检测欺诈点击,构建了多种充分结合广告点击与时间属性关系的特征,并提出了一种点击欺诈检测的集成学习框架——CAT-RFE集成学习框架。CAT-RFE集成学习框架包含3个部分:基分类器、递归特征消除(RFE,recursive feature elimination)和voting集成学习。其中,将适用于类别特征的梯度提升模型——CatBoost(categorical boosting)作为基分类器;RFE是基于贪心策略的特征选择方法,可在多组特征中选出较好的特征组合;Voting集成学习是采用投票的方式将多个基分类器的结果进行组合的学习方法。该框架通过CatBoost和RFE在特征空间中获取多组较优的特征组合,再在这些特征组合下的训练结果通过voting进行集成,获得集成的点击欺诈检测结果。该框架采用了相同的基分类器和集成学习方法,不仅克服了差异较大的分类器相互制约而导致集成结果不理想的问题,也克服了RFE在选择特征时容易陷入局部最优解的问题,具备更好的检测能力。在实际互联网点击欺诈数据集上的性能评估和对比实验结果显示,CAT-RFE集成学习框架的点击欺诈检测能力超过了CatBoost模型、CatBoost和RFE组合的模型以及其他机器学习模型,证明该框架具备良好的竞争力。该框架为互联网广告点击欺诈检测提供一种可行的解决方案。     

网络广告策略研究

网络是自电视发明后诞生的新兴媒体,伴随着国际互联网的逐渐扩大与网络操作技术的日趋完善,电子商务和网络营销迅速崛起,网络广告投放的步伐也随之加快。各种网络广告通过图文信息的视觉整合,不断推出令人赏心悦目的界面效果,给广大浏览者以强烈的视觉冲击和吸引力。网络广告的优势也越来越被广告主所看好。网络广告也由于其交互直接、反馈及时、覆盖面广、无时空差异、针对性强、便于统计、费用低廉等继电视、广播、报纸、杂志和户外广告之后的又一强势媒体。     

Stochastic Budget Optimization in Internet Advertising

Internet advertising is a sophisticated game in which the many advertisers “play” to optimize their return on investment. There are many “targets” for the advertisements, and each “target” has a collection of games with a potentially different set of players involved. In this paper, we study the problem of how advertisers allocate their budget across these “targets”. In particular, we focus on formulating their best response strategy as an optimization problem. Advertisers have a set of keywords (“targets”) and some stochastic information about the future, namely a probability distribution over scenarios of cost vs click combinations. This summarizes the potential states of the world assuming that the strategies of other players are fixed. Then, the best response can be abstracted as stochastic budget optimization problems to figure out how to spread a given budget across these keywords to maximize the expected number of clicks. We present the first known non-trivial poly-logarithmic approximation for these problems as well as the first known hardness results of getting better than logarithmic approximation ratios in the various parameters involved. We also identify several special cases of these problems of practical interest, such as with fixed number of scenarios or with polynomial-sized parameters related to cost, which are solvable either in polynomial time or with improved approximation ratios. Stochastic budget optimization with scenarios has sophisticated technical structure. Our approximation and hardness results come from relating these problems to a special type of (0/1, bipartite) quadratic programs inherent in them. Our research answers some open problems raised by the authors (in Algorithmica, 58(4):1022–1044, 2010).     

基于集成特征选择的点击欺诈检测方法

网络在线广告中以套取广告费为目的的点击欺诈已经严重影响了网络广告的稳定发展。从FDMA2012竞赛提供的欺诈发布商检测的真实数据集出发，针对冗余特征会降低训练效率以及不平衡数据会使决策边界发生偏倚的问题，提出了一种基于集成特征选择的网络在线广告点击欺诈检测方法。采用Bagging方法和合成少数类过采样技术（Synthetic Minority Oversampling Technique，SMOTE）相结合的方法将多数的正常点击广告发布商样本与少数的欺诈点击广告发布商样本构造为多个袋装子集，利用基于相关性度量的特征选择算法对每个袋装子集中筛选出特征子集，设置阈值得到特征合集，利用随机森林算法构建点击欺诈检测模型。实验结果表明该方法能够有效识别出实施欺诈点击行为的非法发布商，达到网络在线广告中点击欺诈检测的要求。     

Representative reviewers for Internet social media

Our many various relationships with persons from home, work and school give rise to our social networks. In a social network, people receive, provide, and pass a great deal of information. In this process, we often observe that certain individuals have especially strong influences on others. We call these highly influential people opinion leaders. Since the late 20th century, the number of Internet users has increased rapidly, and a huge number of people now interact with each other in online social networks. In this way, the Web community has become similar to real-world society. Internet users receive information not only from the mass media, but also from opinion leaders. For example, online articles posted by influential bloggers are often used as marketing tools or political advertisements, due to their huge influence on other users. Therefore, it is important and useful to identify the influential users in an online society. We thus propose a simple yet reliable algorithm that identifies opinion leaders in a cyber social network. In this paper, we first describe our algorithm for identifying influential users in an online society. We then demonstrate the validity of the selection of representative reviewers using the Yahoo! music and GroupLens movie databases and performing 10-fold cross-validation and z-tests.     

交互式多媒体网络广告的设计原则

投放交互式多媒体网络广告的目的是通过引起受众的购买欲望,从而销售商品。为此,在设计和制作广告过程中,广告主和广告设计者需要运用好的营销理念指导,以受众体验为基点,遵循广告内容的连贯性和相关性原则,个性化与一致性原则,功能成本控制原则,并结合适当的交互技术。     

基于集成模型的移动应用广告转化率预测

移动应用广告是互联网广告市场中一种主动的广告形式,它能够分析用户的兴趣爱好,并投其所好,精准投放广告,从而提高用户体验,为广告平台与广告主带来巨大的收益.因此,预测移动应用广告的转化率已成为一个非常重要的研究方向.本文以逻辑回归和两个梯度提升树模型为基础,使用堆叠和平均的集成思想,提出了两种集成模型--SXL和BLLX模型,解决了传统预测模型能力有限,无法精准预测转化率的问题.在腾讯2017社交广告比赛的数据集上的实验结果表明,SXL和BLLX两种模型能够有效地提高广告转化率的预测结果.