Rank-LIME: Local Model-Agnostic Feature Attribution for Learning to Rank

TL;DR

Rank-LIME extends the LIME explanation method to the learning to rank task, providing local, model-agnostic, additive feature attributions for ranked lists, and demonstrates superior fidelity and ranking explanation quality on MS MARCO datasets.

Contribution

This work introduces Rank-LIME, a novel explanation method specifically designed for learning to rank, utilizing correlation-based perturbations and new evaluation metrics.

Findings

Rank-LIME outperforms existing explanation algorithms in fidelity.

Rank-LIME provides effective explanations for ranked lists.

The method is evaluated on MS MARCO datasets.

Abstract

Understanding why a model makes certain predictions is crucial when adapting it for real world decision making. LIME is a popular model-agnostic feature attribution method for the tasks of classification and regression. However, the task of learning to rank in information retrieval is more complex in comparison with either classification or regression. In this work, we extend LIME to propose Rank-LIME, a model-agnostic, local, post-hoc linear feature attribution method for the task of learning to rank that generates explanations for ranked lists. We employ novel correlation-based perturbations, differentiable ranking loss functions and introduce new metrics to evaluate ranking based additive feature attribution models. We compare Rank-LIME with a variety of competing systems, with models trained on the MS MARCO datasets and observe that Rank-LIME outperforms existing explanation algorithms in terms of Model Fidelity and Explain-NDCG. With this we propose one of the first algorithms to generate additive feature attributions for explaining ranked lists.