Building Interpretable Interaction Trees for Deep NLP Models

Authors

  • Die Zhang Shanghai Jiao Tong University
  • Hao Zhang Shanghai Jiao Tong University
  • Huilin Zhou Shanghai Jiao Tong University
  • Xiaoyi Bao Shanghai Jiao Tong University
  • Da Huo Shanghai Jiao Tong University
  • Ruizhao Chen Shanghai Jiao Tong University
  • Xu Cheng Shanghai Jiao Tong University
  • Mengyue Wu Shanghai Jiao Tong University
  • Quanshi Zhang Shanghai Jiao Tong University

DOI:

https://doi.org/10.1609/aaai.v35i16.17685

Keywords:

Interpretaility & Analysis of NLP Models

Abstract

This paper proposes a method to disentangle and quantify interactions among words that are encoded inside a DNN for natural language processing. We construct a tree to encode salient interactions extracted by the DNN. Six metrics are proposed to analyze properties of interactions between constituents in a sentence. The interaction is defined based on Shapley values of words, which are considered as an unbiased estimation of word contributions to the network prediction. Our method is used to quantify word interactions encoded inside the BERT, ELMo, LSTM, CNN, and Transformer networks. Experimental results have provided a new perspective to understand these DNNs, and have demonstrated the effectiveness of our method.

Downloads

Published

2021-05-18

How to Cite

Zhang, D., Zhang, H., Zhou, H., Bao, X., Huo, D., Chen, R., Cheng, X., Wu, M., & Zhang, Q. (2021). Building Interpretable Interaction Trees for Deep NLP Models. Proceedings of the AAAI Conference on Artificial Intelligence, 35(16), 14328-14337. https://doi.org/10.1609/aaai.v35i16.17685

Issue

Section

AAAI Technical Track on Speech and Natural Language Processing III