Extractive Adversarial Networks: High-Recall Explanations for Identifying Personal Attacks in Social Media Posts
arXiv:1809.01499
Abstract
We introduce an adversarial method for producing high-recall explanations of neural text classifier decisions. Building on an existing architecture for extractive explanations via hard attention, we add an adversarial layer which scans the residual of the attention for remaining predictive signal. Motivated by the important domain of detecting personal attacks in social media comments, we additionally demonstrate the importance of manually setting a semantically appropriate `default' behavior for the model by explicitly manipulating its bias term. We develop a validation set of human-annotated personal attacks to evaluate the impact of these changes.
arXiv commentAccepted to EMNLP 2018 Code and data available at https://github.com/shcarton/rcnn
References in corpus (8)
- Adam: A Method for Stochastic Optimization
- Towards A Rigorous Science of Interpretable Machine Learning
- Self-Attention Generative Adversarial Networks
- Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)
- Understanding Neural Networks through Representation Erasure
- Generating Natural Adversarial Examples
- SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient
- Rationalization: A Neural Machine Translation Approach to Generating Natural Language Explanations