方法对比

	簇随机 A/B 测试	自适应A/B测试
领域	实验设计	实验设计
方法族	Process / pipeline	Process / pipeline
起源年份	2010s (digital platforms); cluster RCT roots date to the 1970s–1980s	1952 (Robbins); applied to A/B testing from ~2010s onward
提出者	Developed from cluster randomized trial methodology; popularized in digital experimentation by researchers at Facebook, LinkedIn, and Microsoft Research (2010s)	Herbert Robbins (bandit framework); Thompson Sampling formalized by William R. Thompson
类型	Experimental design	Adaptive experimental design
开创性文献	Ugander, J., Karrer, B., Backstrom, L., & Kleinberg, J. (2013). Graph cluster randomization: Network exposure to multiple universes. Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 329–337.	Russo, D., Van Roy, B., Kazerouni, A., Osband, I., & Wen, Z. (2018). A Tutorial on Thompson Sampling. Foundations and Trends in Machine Learning, 11(1), 1–96.
别名	cluster A/B test, group-randomized A/B test, network A/B test, cluster-level split test	adaptive AB test, bandit A/B test, multi-armed bandit testing, online adaptive experiment
相关	6	6
摘要	A cluster randomized A/B test is an experimental design in which intact groups (clusters) — such as cities, schools, social network communities, or app user segments — are randomly assigned as whole units to either the treatment (A) or control (B) condition, rather than randomizing individual users or subjects. This approach is used when treatment effects would spill over between individuals if individual-level randomization were applied, or when the intervention must be delivered at the group level.	An Adaptive A/B test is an experimental design that dynamically reallocates traffic or participants toward better-performing variants during the experiment itself, rather than holding allocations fixed until the end. Drawing on multi-armed bandit algorithms such as Thompson Sampling or Upper Confidence Bound (UCB), it balances the exploration of uncertain variants with the exploitation of those already showing superior performance, typically yielding higher aggregate outcomes while still producing valid inferential conclusions.