Modeling historical user behavior for enhanced ad targeting

在线广告是许多基于网络的企业的主要收入来源之一。第三方在《环球邮报》的网站上宣传他们的产品和服务。目前，该系统将所有广告、用户需求和行为视为同等重要。因此，在不考虑未来用户行为的情况下应用标准广告费率。Globe和Mail希望了解历史用户行为以及他们与其网站上的广告的互动，以预测客户的购买行为。这样的预测将加强广告定向，并允许他们收取更高的广告费率。以前的购物行为对未来的购物行为有显著影响。了解和预测游客在本次访问中的购买行为具有重要意义。与之前的研究/模型预测访问结束时的购买行为不同，能够在访问过程中早期预测购买行为是很重要的。在这个项目中，我们的目标是建立一个预测模型，预测整个点击流的购买行为。考虑到预测整个访问者点击流的在线购买行为是一个相对较新的领域，这个项目将为这个主题提供更多的洞察力。虽然传统线下门店的消费行为有很大的科学基础，但它与线上的消费行为有很多不同之处。在这个项目中，我们将使用访问级聚合点击流数据和页面浏览级分类点击流数据来整合更详细的信息，以模拟整个点击流的购买行为。

Online advertising is one of the main revenue streams of many web based businesses. Third parties advertise their products and services on Globe and Mail's web site. Currently the system treats all the ads and user needs and behavior as of equal importance. Therefore a standard ad rate is applied without considering the future user behavior. Globe and Mail would like to understand historical user behavior and their interactions with ads on their web site to predict the purchase behavior of their customers. Such a prediction would enhance ad targeting and allow them to charge higher ad rates.Previous shopping behavior has significant effect on future shopping behavior. Understanding and predicting visitor's purchasing behavior during the current visit is of great importance. It is important to be able to predict purchasing behavior early during the visiting process, in contrast to previous research/ models that predicted purchase behavior at the end of the visit. In this project we aim to build a prediction model that predicts the purchase behavior throughout the clickstream. Given that predicting online purchasing behavior throughout the visitor's clickstream is a relatively new area, this project will provide more insight into this topic. Although consumer behavior in traditional offline stores has a large scientific base, it is in many ways different from online consumer behavior. In this project we will use visit level aggregated clickstream data, and pageview level disaggregated clickstream data to incorporate more detailed information to model purchasing behavior throughout the clickstream.