Language Modelling on Explanatory sentences
65BLEUBERT-TransCONV (+ addition QKV)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BERT-TransCONV (+ addition QKV)Backbone=BERT-TransCONV, Injection Method=addition, Injection Target=QKV2026.01 | 65 | 0.35 | 94 | 0.62 | 1.85 | |
| BERT-GraphSAGE (+ addition QKV)Backbone=BERT-GraphSAGE, Injection Method=addition, Injection Target=QKV2026.01 | 63 | 0.31 | 93 | 0.65 | 1.91 | |
| BERT-GraphSAGE (+ addition Q)Backbone=BERT-GraphSAGE, Injection Method=addition, Injection Target=Q2026.01 | 60 | 0.22 | 92 | 0.74 | 2.09 | |
| BERT-TransCONV (+ addition Q)Backbone=BERT-TransCONV, Injection Method=addition, Injection Target=Q2026.01 | 59 | 0.18 | 92 | 0.76 | 2.13 | |
| BERT-GraphSAGE (+ fusion Q)Backbone=BERT-GraphSAGE, Injection Method=fusion, Injection Target=Q2026.01 | 55 | 0.03 | 91 | 0.9 | 2.45 | |
| BERT-TransCONV (+ fusion Q)Backbone=BERT-TransCONV, Injection Method=fusion, Injection Target=Q2026.01 | 53 | -0.02 | 90 | 0.98 | 2.66 | |
| SiamEncoding Setup=memory injection2026.01 | 49 | -0.15 | 88 | 0.94 | 2.55 | |
| BERT-GraphSAGE (+ fusion QKV)Backbone=BERT-GraphSAGE, Injection Method=fusion, Injection Target=QKV2026.01 | 46 | -0.23 | 88 | 1.1 | 3 | |
| BERT-TransCONV (+ fusion QKV)Backbone=BERT-TransCONV, Injection Method=fusion, Injection Target=QKV2026.01 | 46 | -0.16 | 88 | 1.13 | 3.09 | |
| Optimus (GraphEncoder + GraphSAGE)Backbone=GraphEncoder2026.01 | 45 | -0.28 | 87 | 1 | 2.71 | |
| LSTMEncoding Setup=memory injection2026.01 | 41 | -0.41 | 85 | 1.04 | 2.82 | |
| AAE(768)Latent Size=7682026.01 | 35 | -0.95 | 80 | 3.35 | 28.5 | |
| Optimus(768)Latent Size=7682026.01 | 35 | -0.59 | 83 | 0.98 | 2.66 | |
| LAAE(768)Latent Size=7682026.01 | 26 | -1.07 | 78 | 3.71 | 40.85 | |
| VGAEEncoding Setup=memory injection2026.01 | 26 | -0.91 | 78 | 1.14 | 2.55 | |
| DAAE(768)Latent Size=7682026.01 | 22 | -1.26 | 76 | 4 | 54.59 | |
| Optimus (GraphEncoder + TransCONV)Backbone=GraphEncoder2026.01 | 17 | -1.16 | 75 | 1.21 | 3.35 | |
| Optimus (GraphEncoder + GCN)Backbone=GraphEncoder2026.01 | 15 | -1.19 | 75 | 1.24 | 3.45 | |
| β-VAE(768)Latent Size=7682026.01 | 6 | -1.14 | 77 | 3.69 | 40.04 |