Corpus-level and Concept-based Explanations for Interpretable Document Classification

Using attention weights to identify information that is important for models’ decision making is a popular approach to interpret attention-based neural networks. This is commonly realized in practi...