dfki-nlp / language-attributions

Learning Explanations from Language Data

language-attributions

by David Harbecke, Robert Schwarzenberg and Christoph Alt.

Code accompanying the paper

@inproceedings{harbecke_la_2018,
  title = {Learning Explanations from Language Data},
  booktitle = {Proceedings of the EMNLP 2018 Workshop on Analyzing and Interpreting Neural Networks for NLP},
  author = {Harbecke, David and Schwarzenberg, Robert and Alt, Christoph},
  location = {Brussels, Belgium},
  year = {2018}
  }

Applying the PatternAttribution approach by

PJ Kindermans, KT Schütt, M Alber, KR Müller, D Erhan, B Kim, S Dähne. Learning how to explain neural networks: PatternNet and PatternAttribution International Conference on Learning Representations (ICLR), 2018

to language.

Our implementation uses their toolbox.

Here are contributions to a negative sentiment classification that our method retrieved from a CNN sentiment classifier:

The review is taken from the Amazon Review Polarity dataset on which we also trained the sentiment model.

How to run experiments:

• install requirements.txt
• install english spacy model (python -m spacy download en)
• set values in config.INI (might not be necessary)
• run run_experiments.py