Coalition Game Theory In Attribution Modeling
研究了用联盟博弈论中的Shapley值进行广告归因建模，提出一种可扩展的近似方法，解决传统Shapley值计算量大的问题，同时保持可解释性，帮助广告主公平分配转化功劳。
<h3>ABSTRACT</h3> Attribution modeling (AM) has a crucial role in measuring the impact of advertising inputs in driving actions (clicks, conversions, purchases, homepage visits, etc.). A misattributing attribution model, such as last touch, allows publishers to ride freely on others9 efforts. This, in turn, powers futile optimizations with no realized performance lift. Shapley value and logistic regression stand out as reliable attribution models with a reputation across-industry verticals. AM using coalitional game theory—Shapley values—can distribute fairly both gains and costs to inputs, with unequal contributions, working together. AM using Shapley values, however, faces a scalability challenge for most practical applications. Notwithstanding, existing scalable AM methods not only lack interpretability but also blur the contrast between efficiency and contribution. This study demonstrates a scalable way to approximate Shapley values, mainly with successive orders of probabilistic models, which also provide additional insights into the efficiency and contribution of interacting advertising inputs.