Topic Modeling on 20NG
0.24NPMIProdLDA VAE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ProdLDA VAE# topics=502017.03 | 0.24 | — | |
| ProdLDA VAE# topics=2002017.03 | 0.19 | — | |
| LDA Collapsed Gibbs# topics=502017.03 | 0.17 | — | |
| S2WTMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.167 | 0.723 | |
| DVAE-TMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.155 | 0.748 | |
| DVAE-RSVI-TMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.146 | 0.75 | |
| LDA Collapsed Gibbs# topics=2002017.03 | 0.14 | — | |
| NMFTopic count (K)=20, Median over 5 runs=true2025.07 | 0.118 | 0.648 | |
| PRISMAggregation=mean (std) over three topic settings2026.03 | 0.1168 | 0.6592 | |
| LDA VAE (AVITM)# topics=502017.03 | 0.11 | — | |
| LDA VAE (AVITM)# topics=2002017.03 | 0.11 | — | |
| LDA DMFVI# topics=502017.03 | 0.11 | — | |
| ProdLDATopic count (K)=20, Median over 5 runs=true2025.07 | 0.107 | 0.66 | |
| CombinedTMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.107 | 0.655 | |
| ZeroshotTMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.103 | 0.653 | |
| MALLETAggregation=mean (std) over three topic settings2026.03 | 0.1018 | 0.6322 | |
| LDATopic count (K)=20, Median over 5 runs=true2025.07 | 0.092 | 0.599 | |
| BERTopicAggregation=mean (std) over three topic settings2026.03 | 0.0887 | 0.5557 | |
| NVDM# topics=502017.03 | 0.08 | — | |
| C-Top2VecAggregation=mean (std) over three topic settings2026.03 | 0.0682 | 0.5632 | |
| ETMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.066 | 0.564 | |
| LDA DMFVI# topics=2002017.03 | 0.06 | — | |
| NVDM# topics=2002017.03 | 0.06 | — | |
| ProdLDAAggregation=mean (std) over three topic settings2026.03 | 0.0569 | 0.5976 | |
| NMFAggregation=mean (std) over three topic settings2026.03 | 0.055 | 0.5497 | |
| WTMTopic count (K)=20, Median over 5 runs=true2025.07 | 0.046 | 0.505 | |
| LDAAggregation=mean (std) over three topic settings2026.03 | 0.0457 | 0.5321 | |
| vONTTopic count (K)=20, Median over 5 runs=true2025.07 | 0.045 | 0.505 | |
| ETMAggregation=mean (std) over three topic settings2026.03 | 0.0433 | 0.5242 | |
| NeuralLDAAggregation=mean (std) over three topic settings2026.03 | 0.0425 | 0.5339 | |
| FASTopicAggregation=mean (std) over three topic settings2026.03 | 0.0223 | 0.5744 | |
| LSITopic count (K)=20, Median over 5 runs=true2025.07 | 0.006 | 0.457 | |
| ECRTMTopic count (K)=20, Median over 5 runs=true2025.07 | -0.089 | 0.416 |