AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

Estimation of a Low-rank Topic-Based Model for Information Cascades

Ming Yu, Varun Gupta, Mladen Kolar

JMLR 2020 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Mutual information estimation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use the low-rank topic model (influence and receptivity matrices) instead of NetRate's single diffusion matrixNetRate · MemeTracker test cascadesAIC (test): 5.08 x 10^5 -> 2.38 x 10^5 (vs TopicCascade); NetRate 2.60 x 10^5 -> 2.38 x 10^5Improved
use the low-rank topic model with 2pK parameters instead of full TopicCascade with p^2K parametersTopicCascade · MemeTracker test cascadesnumber of parameters: 2500000 -> 10000Improved
use the low-rank model instead of TopicCascadeTopicCascade · MemeTracker test cascadestest negative log-likelihood: 81.8 -> 82.3No clear effect
use the low-rank model instead of NetRate and TopicCascade (train vs test gap)NetRate / TopicCascade · MemeTrackertrain negative log-likelihood: train 68.5 -> 80.3 vs test 81.1 -> 82.3 (NetRate); 62.5 vs 81.8 (TopicCascade)Improved
use the low-rank model instead of TopicCascade at large scale (xi=8)TopicCascade · synthetic running timerunning time (s): 1310.7 -> 454.9Improved
use NetRate (ignores topics) instead of the proposed modelproposed model · synthetic running timerunning time (s): 454.9 -> 211.0 (NetRate is faster)Hurt
use g2 (l2 column norm) instead of g1 (l1) regularization at small sample size (ablation)Algorithm 1 with g1 · synthetic, p=50estimation errorHurt
use g2 (l2 column norm) instead of g1 regularization at large sample size (ablation)Algorithm 1 with g1 · synthetic, p=50estimation errorImproved
reduce dimension to p=10, K=6 and increase sample size (ablation)TopicCascade vs proposed model · synthetic small pestimation errorImproved
use small p=10, K=6 (NetRate vs proposed model)NetRate · synthetic small ptest negative log-likelihoodMixed
use a larger regularization parameter (20 times the optimal) to improve F1 on network recoveryproposed model · synthetic p=50F1 scoreImproved
use a denser graph (5-6 topics per row instead of 2-3) (ablation)proposed model vs baselines · synthetic dense graphnegative log-likelihood and estimation errorNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy