During this webinar, you will learn how to predict the chance of converting or converting odds for keywords with little or no historical data, in Pay-Per-Click programs such as Google, Yahoo and Bing (Microsoft). We will discuss text mining techniques, a rule engine, KPI selection and predictive modeling (logistic regression, decision trees, naïve Bayes or hybrid models) to score bid keywords with no or little historical data, such as new keywords added to existing ad groups. Also, a simple parametric keyword bidding algorithm will be discussed. The parametric bidding algorithm relies on keyword scores and – when available and statistically significant – conversion rates.
The scores are built using text mining rules, keyword grouping, and a training set with multiple clients with various conversion rates, and several million bid keywords. Cross-validation will be discussed. The keyword score (more precisely, a function of the score) is used as a proxy for the conversion rate. The predicted ROI, used in the bidding algorithm, is a simple function of the current bid, the score (which in turn is a predictor of the conversion rate, by design) and the revenue per conversion.
Illustration with lift measurements will be provided on a real example, using data from a very large advertiser, focusing on keywords with low click frequency (the long tail).
About Vincent Granville, PhD
Vincent Granville has successfully solved problems for 15 years in data mining, text mining, predictive modeling, business intelligence, technical analysis, keyword and web analytics. Granville is widely recognized as the leading expert in click scoring and web traffic optimization. Over the last ten years, he has worked in real-time credit card fraud detection with Visa, advertising mix optimization with CNET, A/B testing with LowerMyBills, online user experience with Wells Fargo, search intelligence with InfoSpace, click fraud detection with major search engines and large advertising clients, as well as statistical litigation.
Granville was formerly Chief Science Officer at Authenticlick, where he developed patent pending technology. Most recently, he successfully launched DataShaping and AnalyticBridge, the largest social network for analytic professionals, with 20,000 members. He is a former post-doctorate of Cambridge University and the National Institute of Statistical Sciences. He was among the finalists at the Wharton School Business Plan Competition and at the Belgian Mathematical Olympiads. Granville has published 40 papers in statistical journals and is an invited speaker at international conferences. He also developed a new data mining technology known as hidden decision trees.
