Data Portal @ linkeddatafragments.org

Matches in ScholarlyData for { ?s ?p We present the experiences from building a web-scale user modeling platform for optimizing display advertising targeting at Yahoo!. Unlike the many popular advertising methodologies optimizing for user clicks in predefined demographics and interest categories, the platform described in this paper optimizes for maximizing purchase activities or transactions, called \emph{conversions} by building a per-campaign model. Conversions directly translate to advertiser's revenue, and thus provide the most relevant metrics of return on advertising investment. We focus on two major challenges: how to efficiently process histories of billions of users on daily basis, and how to build per-campaign conversion models given the extremely low conversion rates (compared to clicks in traditional setting). We first present mechanisms for building web-scale user profiles in a daily incremental fashion. Second, we show how to reduce the latency by in-memory processing of billions of user records. Finally, we discuss a technique for scaling in the number of models by efficient labeling technique that allows for sharing negative training examples across multiple campaigns. As our platform showed superiority over existing state-of-the-art targeting methods, it is currently deployed in production, allowing thousands of Yahoo! advertisers to optimize the performance of their campaigns.Based on a rigorous offline empirical evaluation over 1776 advertising campaigns, we explore different improvements for our platform such as evaluating different techniques for feature weighting, such as weighting based on user activity counts, and recency-based feature weighting. Additionally, we develop a robust feature selection mechanism that optimizes running times without affecting modeling performance.. }

• 20 abstract "We present the experiences from building a web-scale user modeling platform for optimizing display advertising targeting at Yahoo!. Unlike the many popular advertising methodologies optimizing for user clicks in predefined demographics and interest categories, the platform described in this paper optimizes for maximizing purchase activities or transactions, called \emph{conversions} by building a per-campaign model. Conversions directly translate to advertiser's revenue, and thus provide the most relevant metrics of return on advertising investment. We focus on two major challenges: how to efficiently process histories of billions of users on daily basis, and how to build per-campaign conversion models given the extremely low conversion rates (compared to clicks in traditional setting). We first present mechanisms for building web-scale user profiles in a daily incremental fashion. Second, we show how to reduce the latency by in-memory processing of billions of user records. Finally, we discuss a technique for scaling in the number of models by efficient labeling technique that allows for sharing negative training examples across multiple campaigns. As our platform showed superiority over existing state-of-the-art targeting methods, it is currently deployed in production, allowing thousands of Yahoo! advertisers to optimize the performance of their campaigns.Based on a rigorous offline empirical evaluation over 1776 advertising campaigns, we explore different improvements for our platform such as evaluating different techniques for feature weighting, such as weighting based on user activity counts, and recency-based feature weighting. Additionally, we develop a robust feature selection mechanism that optimizes running times without affecting modeling performance.".