BriefGPT.xyz
Aug, 2018
神经响应生成模型为什么更倾向于使用通用回复?
Why Do Neural Response Generation Models Prefer Universal Replies?
HTML
PDF
Bowen Wu, Nan Jiang, Zhifeng Gao, Suke Li, Wenge Rong...
TL;DR
该研究分析了基于序列到序列学习的生成回复任务的神经模型容易产生短而通用回复的问题,并通过分解黑匣子,详细分析了概率极限问题并提出了最大间隔排名正则化方法来避免模型偏向于这些回复,并通过实证实验验证了该方法的有效性。
Abstract
Recent advances in
sequence-to-sequence learning
reveal a purely data-driven approach to the
response generation
task. Despite its diverse applications, existing
→