节点文献
基于门户网站的精准广告系统架构设计
A Portal-based Precision Advertising System Architecture Design
【作者】 王茜；
【作者基本信息】 天津大学 ， 物流工程， 2014， 硕士
【摘要】 在线广告作为互联网最主要的盈利模式之一，支撑着互联网行业的发展。相比传统的线下广告，在线广告可根据人口统计信息等因素精准投送，其效果也可通过相关统计工具衡量出，因而受到各大广告主青睐。纵观在线广告发展史，如何充分利用剩余流量并提高广告主、媒体及用户收益是在线广告目前所要解决的核心问题；作为一个需求驱动的互联网产品，在不同的使用场景下，系统架构及算法选择也各有不同。本文将以充分利用门户网站剩余流量为根本，设计一个简单的精准广告投放系统，系统包含五个模块：Adserver、Sessionlog、Targeting、Adindex及Adexchange；共支持三种广告的推送逻辑：定向广告、相似推荐广告及RTB广告，涵盖了精准广告系统的大部分核心模块。为解决定向广告检索问题及长Query广告匹配问题，系统将选择DNF及Weak AND作为广告检索模块的核心算法，并使用流归并的方法，针对广告条件属性多样化及用户标签属性多样化，对DNF算法做两部分改进，从而更好地解决门户网站计算广告的核心问题，即在不同的页面，为不同的用户投放广告，优化整体投入产出比。本文最大的独创性在于对DNF算法的改进，使用流归并的方法，解决了用户标签属性的多值“与”关系检索，以及广告条件属性的多值“与”关系索引建立；且两部分改进可以无冲突地使用于同一系统中，极大的方便了广告主对定向广告条件的设定。
【Abstract】 Online advertising is one of the Internet’s main profit models. It supports thedevelopment of the Internet industry. Compared to traditional advertising, mostadvertisers prefer to online advertising. Online advertising can be precision pushed tothe Internet users based on demographic and other factors. Meanwhile, the statisticaltools can measure the effect. Throughout the history of online advertising, how tomake full use of the remaining traffic and improve advertisers, media and users gainis currently the core issues to be addressed. As a demand-driven Internet product, thesystem architecture and algorithm selection is also diverse under different usagescenarios.To take full advantage of the remaining traffic this article designs a simple portalprecision advertising system. The system consists of five modules: Ad server, Sessionlog, Targeting, Ad index and Ad exchange. It supports three advertising push logicincluding Targeted Advertising, similar recommendations and Ad Exchange andcovered most of the core modules of accurate advertising system. To solve theproblem of qualitative ad retrieval and long-Query ad matching problem, the systemwill choose DNF and Weak AND as the core algorithm in Ad index. Meanwhile, thisarticle using the stream merge method improves the DNF algorithm. To better meetthe needs of diverse attributes of advertisers’ demand and the diverse attributes of usertab in index. Therefore, solving the advertising portal computing core issues that pushbest ads for different users on different pages and optimizing the overall input-outputratio.The DNF algorithm improvements are the greatest originality of this article.Using stream merge method, the improvements solve the "and" relationship ofmulti-value user tag attribute search and the "and" relationship of multi-valuedattribute indexing. Using the two parts improvements in the same system bringsconvenience to the Targeting Criteria.