BriefGPT.xyz
Oct, 2019
PLATO：基于离散潜变量的预训练对话生成模型
PLATO: Pre-trained Dialogue Generation Model with Discrete Latent Variable
HTML
PDF
Siqi Bao, Huang He, Fan Wang, Hua Wu
TL;DR
本文提出了基于预训练模型的对话生成框架，采用灵活的注意力机制和离散的潜在变量，解决了响应生成中存在的一对多映射问题，并设计了两种互补的任务对话响应生成和潜在动作识别。实验结果表明，该框架在三个公开数据集上验证了其优越性。
Abstract
pre-training
models have been proved effective for a wide range of
natural language processing
tasks. Inspired by this, we propose a novel
dialog
→