TIPS: A text interaction evaluation metric for learning model interpretation

Zhenyu Nie, Zheng Xiao, Tao Wang, Anthony Theodore Chronopoulos, Razvan Andonie, Amir Mosavi

Abstract: Highlights•Proposing a novel metric TIPS to evaluate interaction-based interpretation methods.•Eliminating all non-target interactions from the text to retain the target one.•Using relative interaction scores mitigates interference of text interaction.•TIPS provides an unbiased evaluation and enhances the confidence of explanations.