dialog context c:前k-1個(gè)utterances，conversational floor（1或0）,meta features(topic)
latent variable z: capture distribution of valid responses
x: response utterance
y:linguistic features(knowledge-guided CVAE)
p(z|c):prior network
p(x|z,c): response decoder惕耕，用q(x|z,c) recognition network來(lái)模擬
生成過(guò)程：sample z, generate x如圖c所示
訓(xùn)練過(guò)程辜妓，如圖b所示蹲缠，通過(guò)max L目標(biāo)函數(shù)得到q p的兩個(gè)參數(shù)，從而得到z的分布粥诫。再由z生成response
The neural network architectures for the baseline and the proposed CVAE/kgCVAE models. The dashed blue connections only appear in kgCVAE
