A deterministic strongly polynomial algorithm for matrix scaling and approximate permanents
STOC '98 Proceedings of the thirtieth annual ACM symposium on Theory of computing
Proceedings of the 9th international World Wide Web conference on Computer networks : the international journal of computer and telecommunications netowrking
Approximating Aggregate Queries about Web Pages via Random Walks
VLDB '00 Proceedings of the 26th International Conference on Very Large Data Bases
Fastest Mixing Markov Chain on a Graph
SIAM Review
Group formation in large social networks: membership, growth, and evolution
Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining
The dynamics of viral marketing
ACM Transactions on the Web (TWEB)
Walking in facebook: a case study of unbiased sampling of OSNs
INFOCOM'10 Proceedings of the 29th conference on Information communications
Walking on a graph with a magnifying glass: stratified sampling via weighted random walks
Proceedings of the ACM SIGMETRICS joint international conference on Measurement and modeling of computer systems
Walking on a graph with a magnifying glass: stratified sampling via weighted random walks
ACM SIGMETRICS Performance Evaluation Review - Performance evaluation review
Framework and algorithms for network bucket testing
Proceedings of the 21st international conference on World Wide Web
Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining
Graph cluster randomization: network exposure to multiple universes
Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining
When web personalization misleads bucket testing
Proceedings of the 1st workshop on User engagement optimization
Hi-index | 0.01 |
Bucket testing, also known as A/B testing, is a practice that is widely used by on-line sites with large audiences: in a simple version of the methodology, one evaluates a new feature on the site by exposing it to a very small fraction of the total user population and measuring its effect on this exposed group. For traditional uses of this technique, uniform independent sampling of the population is often enough to produce an exposed group that can serve as a statistical proxy for the full population. In on-line social network applications, however, one often wishes to perform a more complex test: evaluating a new social feature that will only produce an effect if a user and some number of his or her friends are exposed to it. In this case, independent uniform draws from the population will be unlikely to produce groups that contains users together with their friends, and so the construction of the sample must take the network structure into account. This leads quickly to challenging combinatorial problems, since there is an inherent tension between producing enough correlation to select users and their friends, but also enough uniformity and independence that the selected group is a reasonable sample of the full population. Here we develop an algorithmic framework for bucket testing in a network that addresses these challenges. First we describe a novel walk-based sampling method for producing samples of nodes that are internally well-connected but also approximately uniform over the population. Then we show how a collection of multiple independent subgraphs constructed this way can yield reasonable samples for testing. We demonstrate the effectiveness of our algorithms through computational experiments on large portions of the Facebook network.