Multi-touch attribution of ad models
Multi-touch attribution allows credit to more than one ad - rather than last ad a user sees - especially when multiple media channels, such as search, display, social, mobile and video are involved. Attribution problem thus focuses more on accurate and stable interpretation of influence of each user interaction to final user decision rather than just user classification. [1] first proposes a bivariate metric, one measures variability of estimate, and other measures accuracy of classifying positive and negative users. Then a bagged logistic regression model is developed, which achieves a comparable classification accuracy as a usual logistic regression, but a much more stable estimate of individual advertising channel contributions. An intuitive probabilistic model is proposed to directly quantify attribution of different advertising channels.
[2] presents a causally motivated methodology for conversion attribution in online advertising campaigns. Need for standardization of attribution measurement is discussed and three guiding principles are offered to contribute to this standardization. Stemming from these principles, attribution is positioned as a causal estimation problem and two approximation methods are then proposed as alternatives for when full causal estimation cannot be done. These approximate methods derive from the causal approach and incorporate prior attribution work in cooperative game theory. In cases where causal assumptions are violated, these approximate methods can be interpreted as variable importance measures.
Drawback of rule-based attribution models lies in the fact that rules are not derived from data but only based on simple intuition. With the ever enhanced capability to tracking advertisement and users’ interaction with advertisement, data-driven multi-touch attribution models, which attempt to infer contribution from user interaction data, become an important research direction. [3] proposes such a new model based on survival theory. By adopting a probabilistic framework, one key advantage of proposed model is that it is able to remove presentation biases inherent to most other attribution models. In addition to model the attribution, the proposed model is also able to predict user’s conversion probability.
Budget allocation in online advertising deals with distributing campaign (insertion order) level budgets to different sub-campaigns, which employ different targeting criteria and may perform differently in terms of campaign-level and advertiser RoI [4]. To do this, it is crucial to be able to correctly determine performance of sub-campaigns. This determination is highly related to action-attribution problem, i.e. to be able to find out set of ads, and hence sub-campaigns that provided them to a user, that an action should be attributed to. For this purpose, both last-touch (last ad gets all credit) and multi-touch (many ads share the credit) attribution methodologies are employed.