0

Jointly Optimizing Diversity and Relevance in Neural Response Generation

本文提出了正则化的多任务学习框架SpaceFusion，通过结构化隐变量空间联合优化多样性和相关性。NAACL2019 paper linkcode link