On the Granularity of Explanations in Model Agnostic NLP Interpretability
Current methods for Black-Box NLP interpretability, like LIME or SHAP, are based on altering the text to interpret by removing words and modeling the Black-Box response. In this paper, we outline limitations of this approach when using complex BERT-based classifiers: The word-based sampling produces texts that are out-of-distribution for the classifier and further gives rise to a highdimensional search space, which can't be sufficiently explored when time or computation power is limited. Both of these challenges can be addressed by using segments as elementary building blocks for NLP interpretability. As illustration, we show that the simple choice of sentences greatly improves on both of these challenges. As a consequence, the resulting explainer attains much better fidelity on a benchmark classification task.
WOS:000967751800033
2023-01-01
978-3-031-23617-4
978-3-031-23618-1
Cham
Communications in Computer and Information Science
1752
498
512
REVIEWED
Event name | Event place | Event date |
Grenoble, FRANCE | Sep 19-23, 2022 | |