Generating Plausible Counterfactual Explanations for Deep Transformers in Financial Text Classification

• DOI: 10.18653/v1/2020.coling-main.541
• Publication Date: 2021-01-08T13:58:31Z
• Authors: Linyi Yang; Eoin Kenny; Tin Lok James Ng; Yi Yang; Barry Smyth; Ruihai Dong
• Keywords: Regularization

Corporate mergers and acquisitions (M&A) account for billions of dollars investment globally every year offer an interesting challenging domain artificial intelligence. However, in these highly sensitive domains, it is crucial to not only have a robust/accurate model, but be able generate useful explanations garner user's trust the automated system. Regrettably, recent research regarding eXplainable AI (XAI) financial text classification has received little no attention, many current methods generating textual-based result implausible explanations, which damage To address issues, this paper proposes novel methodology producing plausible counterfactual whilst exploring regularization benefits adversarial training on language models FinTech. Exhaustive quantitative experiments demonstrate that does approach improve model accuracy when compared state-of-the-art human performance, also generates are significantly more based trials.