List your company

Dialogue Generation on PersonaChat (test)

2.31Persona ConsistencyLMEDR
-0.05080.56211.1751.7879Nov 30, 2021Feb 6, 2022Apr 15, 2022Jun 22, 2022Aug 29, 2022Nov 5, 2022Jan 12, 2023

Evaluation results

MethodLinks
LMEDRbackbone=BART-large2023.012.31—————————————3.57—2.94——————
P2BOT2023.012.2—————————————3.51—2.86——————
LIC2023.012.11—————————————3.27—2.69——————
GPT2-D3Backbone=GPT2, Data Manipulation=D3, Training=Fine-tuned2022.040.54815.69——4.1841.4290.18354.6146.4267.3212.2679.80318.20.5572.5322.255———————
GPT2-FILTERBackbone=GPT2, Data Manipulation=Entropy Filter, Training=Fine-tuned2022.040.53716.9——3.7341.3370.17884.576.3527.2632.1489.03116.520.5712.5272.233———————
HumanType=Human Baseline2022.040.531——————5.688.91310.275.25934.966.370.4722.6252.451———————
TRANS-D3Backbone=Transformer, Data Manipulation=D3, Training=From scratch2022.040.51337.3——3.3581.2060.15744.2236.1657.2981.8267.92314.420.4852.3972.172———————
GPT2Backbone=GPT2, Data Manipulation=None, Training=Fine-tuned2022.040.50817.63——3.7611.2780.16934.4856.1877.0292.0118.2615.030.5182.5082.243———————
TRANS-FILTERBackbone=Transformer, Data Manipulation=Entropy Filter, Training=From scratch2022.040.49238.99——2.9461.1010.15634.2836.0337.0881.7967.69614.060.4462.3182.088———————
GPT2-BTBackbone=GPT2, Data Manipulation=Back Translation, Training=Fine-tuned2022.040.45416.96——3.9431.3480.16634.5476.2487.0891.9478.11314.940.5092.4882.259———————
GPT2-CVAEBackbone=GPT2, Data Manipulation=CVAE, Training=Fine-tuned2022.040.42617.16——3.3391.360.15924.2455.6916.491.7486.79912.190.4842.3582.15———————
Transfer Transfo+PEConditioning=Self Persona2021.110.3517.4121.182———————————————————
TRANS-BTBackbone=Transformer, Data Manipulation=Back Translation, Training=From scratch2022.040.3537.92——3.3151.0820.15274.2745.9056.7521.767.10813.390.2892.3372.142———————
Persona GeneratorConditioning=Self Persona2021.110.3417.3121.181.2———————————————————
TRANS-CVAEBackbone=Transformer, Data Manipulation=CVAE, Training=From scratch2022.040.33537.61——3.3121.1910.15333.9745.4516.2671.4595.79511.160.262.3332.111———————
Transfer TransfoConditioning=Self Persona2021.110.3217.7821.580.1———————————————————
TRANSBackbone=Transformer, Data Manipulation=None, Training=From scratch2022.040.30438.28——3.141.1480.14864.0465.4846.2621.6096.29811.710.2352.3032.038———————
Transfer Transfo+PEConditioning=Their Persona2021.110.2818.1521.877.2———————————————————
Persona GeneratorConditioning=Their Persona2021.110.2818.1121.976.8———————————————————
Transfer TransfoConditioning=Their Persona2021.110.2518.3122.376.5———————————————————
Persona ApproximatorConditioning=Self Persona2021.110.2218.3319.873.3———————————————————
Persona ApproximatorConditioning=Their Persona2021.110.1618.5920.471.2———————————————————
GPMNConditioning=Self Persona2021.110.1536.1113.554.9———————————————————
DeepCopyConditioning=Self Persona2021.110.1136.3512.252.7———————————————————
GPMNConditioning=Their Persona2021.110.136.4514.851.4———————————————————
DeepCopyConditioning=Their Persona2021.110.0736.7713.749.6———————————————————
Transfer Transfo w/o personaConditioning=None2021.110.0419.8718.467.3———————————————————
BARTTraining=fine-tuning, Beam search width=52023.06—————————————————9.953.388.696.55——
BART + DialoGPSTraining=fine-tuning, Beam search width=52023.06—————————————————10.974.089.266.7——
DialoFlowTraining=fine-tuning, Beam search width=52023.06—————————————————10.463.039.326.89——
GCNBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————10.113.0114.54—21.4316.46
GCNBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————9.783.1214.47—20.916.59
GCNBackbone=GPT-2, Decoding=greedy2026.06—————————————————6.371.513.06—15.1713.52
HyPEBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————11.123.8415.92—22.6417.5
HyPEBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————10.53.5815.9—22.1817.81
HyPEBackbone=GPT-2, Decoding=greedy2026.06—————————————————7.311.5913.22—17.9414.46
HyPE-baseBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————10.073.1714.53—21.3216.43
HyPE-baseBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————9.863.2915.19—21.2517.13
HyPE-baseBackbone=GPT-2, Decoding=greedy2026.06—————————————————7.281.6813.47—17.0514.16
MeanPoolBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————11.012.0214.82—22.3117.2
MeanPoolBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————9.543.1514.82—20.3917.09
MeanPoolBackbone=GPT-2, Decoding=greedy2026.06—————————————————7.311.5412.47—17.5914.25
ORIGBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————10.22.9415.09—21.5117.11
ORIGBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————9.762.7414.53—21.1116.39
ORIGBackbone=GPT-2, Decoding=greedy2026.06—————————————————5.591.818.67—12.6814.59
Text BaselineBackbone=LLaMA-3.2-3B, Decoding=greedy2026.06—————————————————10.443.7615.29—21.4816.98
Text BaselineBackbone=Qwen2.5-3B, Decoding=greedy2026.06—————————————————10.242.7515.11—21.6417.1
Text BaselineBackbone=GPT-2, Decoding=greedy2026.06—————————————————5.061.058.85—12.6312.11