Estimating conversion rate in display advertising from past erformance data

Abstract

In targeted display advertising, the goal is to identify the best opportunities to display a banner ad to an online user who is most likely to take a desired action such as purchasing a product or signing up for a newsletter. Finding the best ad impression, i.e., the opportunity to show an ad to a user, requires the ability to estimate the probability that the user who sees the ad on his or her browser will take an action, i.e., the user will convert. However, conversion probability estimation is a challenging task since there is extreme data sparsity across different data dimensions and the conversion event occurs rarely. In this paper, we present our approach to conversion rate estimation which relies on utilizing past performance observations along user, publisher and advertiser data hierarchies. More specifically, we model the conversion event at different select hierarchical levels with separate binomial distributions and estimate the distribution parameters individually. Then we demonstrate how we can combine these individual estimators using logistic regression to accurately identify conversion events. In our presentation, we also discuss main practical considerations such as data imbalance, missing data, and output probability calibration, which render this estimation problem more difficult but yet need solving for a real-world implementation of the approach. We provide results from real advertising campaigns to demonstrate the effectiveness of our proposed approach.

User Response Learning for Directly Optimizing Campaign Performance in Display Advertising

“…The user response prediction, such as the click-through rate (CTR) estimation or the conversion rate (CVR) estimation, has become a core research problem in real-time display advertising [14,18,26]. The response prediction is a probability estimation task [19] which models the interest of users towards the content of publishers or the ads, and is used to derive the budget allocation of the advertisers [23].…”

“…Typically, the response prediction problem is formulated as a regression problem with prediction likelihood as the training objective [23,9,1,21]. From the methodology view, linear models such as logistic regression [14] and non-linear models such as tree-based model [10] and factorization machines [19,21] are commonly used. Other variants include Bayesian probit regression [9], FTRFL [24] in factorization machine, and convolutional neural network learning framework [17].…”

“…With the estimated CTR/CVR, the advertisers would be able to assess the value of the impression and perform a bid. The auction theory [7] proves that truthful bidding, i.e., bidding the action value times the action rate, is the optimal strategy in the second price auction [14]. However, with budget and auction volume constraints, the truthful bidding may not be optimal [30].…”

“…Previous studies have specified the bidding strategy as a function b(f θ (x)) mapping from the predicted CTR (or other KPIs) f θ (x) to the bid price [14,22,32]. Essentially, the mapping follows a sequential dependency assumption x → f θ (x) → b proposed by [32,31].…”

“…The common objective in this stage is to make the estimation as accurate as possible by, for instance, minimizing the error between the predictions and ground truth user responses. Second, once we have obtained the user response prediction, we will use it as an input for optimizing the bid based on other considerations including campaign budget, market price, etc [14,32,31,22].…”