http://hdl.handle.net/2451/31634 Search the Channel search http://archive.nyu.edu/simple-search http://hdl.handle.net/2451/31637 Title: Evaluating and Optimizing Online Advertising: Forget the click, butthere are good proxies<br/><br/>Dalessandro, Brian; Hook, Rod; Perlich, Claudia; Provost, Foster<br/><br/>Abstract: A main goal of online display advertising is to drive purchases (etc.)following ad engagement. However, there often are too few purchaseconversions for campaign evaluation and optimization, due to lowconversion rates, cold start periods, and long purchase cycles (e.g.,with brand advertising). This paper presents results across dozens ofexperiments within individual online display advertising campaigns, eachcomparing different 'proxies' for measuring success. Measuring successis critical both for evaluating and comparing different targetingstrategies, and for designing and optimizing the strategies in the firstplace (for example, via predictive modeling). Proxies are necessarybecause data on the actual goals of advertising (e.g., purchasing,increased brand affinity, etc.) often are scarce, missing, orfundamentally difficult or impossible to observe. The paper presents badnews and good news. The most commonly cited and used proxy for successis a click on an advertisement. The bad news is that across a largenumber of campaigns, clicks are not good proxies for evaluation nor foroptimization: buyers do not resemble clickers. The good news is that analternative sort of proxy performs remarkably well: observed visits tothe brand's website. Specifically, predictive models built based onbrand site visits do a remarkably good job of predicting which browserswill purchase. The practical bottom line: evaluating campaigns andoptimizing based on clicks seems wrongheaded; however, there is an easyand attractive alternative|use a well-chosen site visit proxy instead. Thu, 18 Oct 2012 13:43:13 GMT http://hdl.handle.net/2451/31635 Title: Data Science and Prediction<br/><br/>Dhar, Vasant<br/><br/>Abstract: The use of the term 'Data Science' is becoming increasingly common alongwith 'Big Data.' What does Data Science mean? Is there something uniqueabout it? What skills should a 'data scientist' possess to be productivein the emerging digital age characterized by a deluge of data? What arethe implications for business and for scientific inquiry? In this briefmonograph I address these questions from a predictive modeling perspective. Tue, 16 Oct 2012 18:38:31 GMT http://hdl.handle.net/2451/31636 Title: Comparing Context-Aware Recommender Systems in Terms of Accuracy andDiversity: Which Contextual Modeling, Pre-filtering and Post-FilteringMethods Perform the Best<br/><br/>Panniello, Umberto; Tuzhilin, Alexander; Gorgoglione, Michele<br/><br/>Abstract: Although the area of Context-Aware Recommender Systems (CARS) has made asignificant progress over the last several years, the problem ofcomparing various contextual pre-filtering, post-filtering andcontextual modeling methods remained fairly unexplored. In this paper,we address this problem and compare several contextual pre-filtering,post-filtering and contextual modeling methods in terms of the accuracyand diversity of their recommendations to determine which methodsoutperform the others and under which circumstances. To this end, weconsider three major factors affecting performance of CARS methods, suchas the type of the recommendation task, context granularity and the typeof the recommendation data. We show that none of the considered CARSmethods uniformly dominates the others across all of these factors andother experimental settings; but that a certain group of contextualmodeling methods constitutes a reliable &ldquo;best bet&rdquo; whenchoosing a sound CARS approach since they provide a good balance ofaccuracy and diversity of contextual recommendations. Thu, 18 Oct 2012 13:38:28 GMT