TIGTEC : Token Importance Guided TExt Counterfactuals

Milan Bhan (1, 2) , Jean-Noël Vittaut (2) , Nicolas Chesneau (1) , Marie-Jeanne Lesot (2)

1 Ekimetrics
2 LFI - Learning, Fuzzy and Intelligent systems

Counterfactual examples explain a prediction by highlighting changes in an instance that flip the outcome of a classifier. This paper proposes TIGTEC, an efficient and modular method for generating sparse, plausible and diverse counterfactual explanations for textual data. TIGTEC is a text editing heuristic that targets and modifies words with high contribution using local feature importance. A new attention-based local feature importance is proposed. Counterfactual candidates are generated and assessed with a cost function integrating a semantic distance, while the solution space is efficiently explored in a beam search fashion. The conducted experiments show the relevance of TIGTEC in terms of success rate, sparsity, diversity and plausibility. This method can be used in both modelspecific or model-agnostic way, which makes it very convenient for generating counterfactual explanations.

Mots clés: XAI; NLP; Counterfactual examples; Local Feature Importance; Attention

Domaines: Intelligence artificielle [cs.AI]

Milan Bhan, Jean-Noël Vittaut, Nicolas Chesneau, Marie-Jeanne Lesot. TIGTEC : Token Importance Guided TExt Counterfactuals. European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases. ECML PKDD 2023, Sep 2023, Turin, Italy. pp.496-512, ⟨10.1007/978-3-031-43418-1_30⟩. ⟨hal-04311749⟩