ContrXT: Generating contrastive explanations from any text classifier

Lorenzo Malandri, Fabio Mercorio, Mario Mezzanzanica, Navid Nobani, Andrea Seveso

Abstract: Highlights•A method to compute global, t-contrast explanations of text classifiers.•Evaluation on a benchmark dataset and a real-life application with a user-study.•Implementation of the method as an off-the-shelf tool publicly available.•Experimental results showing the effectiveness of the explanation provided.•ContrXT is provided as a service through public REST-API.