Not Logged In

RANCC: Rationalizing Neural Networks via Concept Clustering

We propose a new self-explainable model for Natural Language Processing (NLP) text classification tasks. Our approach constructs explanations concurrently with the formulation of classification predictions. To do so, we extract a rationale from the text, then use it to predict a concept of interest as the final prediction. We provide three types of explanations: 1) rationale extraction, 2) a measure of feature importance, and 3) clustering of concepts. In addition, we show how our model can be compressed without applying complicated compression techniques. We experimentally demonstrate our explainability approach on a number of well-known text classification datasets.

Citation

H. Bashier, M. Kim, R. Goebel. "RANCC: Rationalizing Neural Networks via Concept Clustering". Conference on Computational Linguistics (COLING), pp 3214–3224, December 2020.

Keywords:  
Category: In Conference
Web Links: doi
  ACL

BibTeX

@incollection{Bashier+al:COLING20,
  author = {Housam K. Bashier and Mi-Young Kim and Randy Goebel},
  title = {RANCC: Rationalizing Neural Networks via Concept Clustering},
  Pages = {3214–3224},
  booktitle = {Conference on Computational Linguistics (COLING)},
  year = 2020,
}

Last Updated: January 19, 2021
Submitted by Sabina P

University of Alberta Logo AICML Logo