search

Computational-Content-Analysis-2018 / 16-Feb-3-Bayesian-Poisson-Tensor-Factorization-for-Inferring-Multilateral-Relations-from-Sparse-Dyad

Schein, Aaron, David Blei, John Paisley, Hannah Wallach. 2015. “Bayesian Poisson Tensor Factorization for Inferring Multilateral Relations from Sparse Dyadic Event Counts.” KDD ’15 August 11–13, 2015, Sydney, NSW, Australia.
0 stars 1 forks source link

Basic q... #3

Open Moloq opened 6 years ago

Moloq commented 6 years ago

Not a deep question, but it wasn't very clear to me at all... In the conclusions, they mention that using dyadic events (which sounds to be the same as the orientation reading, or very similar) is not the right approach for multilateral events because of an assumption of independence that isn't valid. Could anyone expand on that? Why would this be the case? Does it affect the validity of other uses of dyadic events?

khan1792 commented 6 years ago

You misunderstand their conclusion. They say that many researchers who use traditional regressions (usually based on Gaussian distribution as I know) think it is biased, but other researchers using Bayesian methods can control the effects from unobserved dependencies.