New research questions the validity of the community-detection algorithm testing model, in which researchers run the algorithm on known data from a real-world network and check to see if their results align with metadata from that network.
"Our research rigorously shows that using metadata as ground truth to validate algorithms is fundamentally problematic and introduces biases without telling us what we really need to know: does my algorithm work?" says the Santa Fe Institute's Daniel Larremore.
He says using metadata for algorithm validation constrains the types of communities that can be validated, which means "we'll only ever get algorithms that solve a small and restricted set of problems."
Larremore and colleagues also demonstrated the futility of developing a universal algorithm for network community-finding, since any algorithm that is outstanding at finding communities in one type of network must be exceptionally poor at finding communities in another.
From Santa Fe Institute
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found