BriefGPT.xyz
Oct, 2022
在文本分类中生成分层解释且不需连接规则
Generating Hierarchical Explanations on Text Classification Without Connecting Rules
HTML
PDF
Yiming Ju, Yuanzhe Zhang, Kang Liu, Jun Zhao
TL;DR
本文提出了一种不需要连接规则的层次化解释生成框架，可以用于生成单词的层次式聚类，以提供高质量的解释，并反映模型决策过程。
Abstract
The opaqueness of
deep nlp models
has motivated the development of methods for interpreting how deep models predict. Recently, work has introduced
hierarchical attribution
, which produces a hierarchical clusterin
→