This website requires JavaScript.

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

Tanya ChowdhuryRazieh RahimiJames Allan
Dec 2022
摘要
Understanding why a model makes certain predictions is crucial when adaptingit for real world decision making. LIME is a popular model-agnostic featureattribution method for the tasks of classification and regression. However, thetask of learning to rank in information retrieval is more complex in comparisonwith either classification or regression. In this work, we extend LIME topropose Rank-LIME, a model-agnostic, local, post-hoc linear feature attributionmethod for the task of learning to rank that generates explanations for rankedlists. We employ novel correlation-based perturbations, differentiable ranking lossfunctions and introduce new metrics to evaluate ranking based additive featureattribution models. We compare Rank-LIME with a variety of competing systems,with models trained on the MS MARCO datasets and observe that Rank-LIMEoutperforms existing explanation algorithms in terms of Model Fidelity andExplain-NDCG. With this we propose one of the first algorithms to generateadditive feature attributions for explaining ranked lists.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答