arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.09197cs.CLcs.AIcs.LG
收藏

时间概念漂移对模型解释的影响

On the Impact of Temporal Concept Drift on Model Explanations

University of Sheffield(谢菲尔德大学)

Zhixue Zhao, George Chrysostomou, Kalina Bontcheva, Nikolaos Aletras

摘要:

自然语言处理中模型预测的解释忠实性通常在来自与训练数据相同的时间分布(即同步设置)的保持数据上进行评估。虽然模型性能通常由于时间变化(即时间概念漂移)而恶化,但目前尚不清楚当目标数据的时间跨度不同于用于训练模型的数据(即异步设置)时,解释忠实性如何受到影响。为此,我们研究了时间变化对模型解释的影响,这些模型解释是由八种特征归因方法和三种选择然后预测模型在六个文本分类任务中提取的。我们的实验表明,(i)在不同的特征归因方法中,忠诚度在时间变化下是不一致的(例如,它会根据方法的不同而减少或增加),基于注意力的方法在数据集上表现出最强大的忠诚度得分;(ii)选择然后预测模型在异步设置中大多是强大的,预测性能只有很小的下降。最后,特征归因方法在用于FRESH(即选择和预测模型)和用于测量充分性/全面性(即事后方法)时显示出冲突的行为,这表明我们需要更强大的指标来评估事后解释的忠实性。

英文摘要:

Explanation faithfulness of model predictions in natural language processing is typically evaluated on held-out data from the same temporal distribution as the training data (i.e. synchronous settings). While model performance often deteriorates due to temporal variation (i.e. temporal concept drift), it is currently unknown how explanation faithfulness is impacted when the time span of the target data is different from the data used to train the model (i.e. asynchronous settings). For this purpose, we examine the impact of temporal variation on model explanations extracted by eight feature attribution methods and three select-then-predict models across six text classification tasks. Our experiments show that (i)faithfulness is not consistent under temporal variations across feature attribution methods (e.g. it decreases or increases depending on the method), with an attention-based method demonstrating the most robust faithfulness scores across datasets; and (ii) select-then-predict models are mostly robust in asynchronous settings with only small degradation in predictive performance. Finally, feature attribution methods show conflicting behavior when used in FRESH (i.e. a select-and-predict model) and for measuring sufficiency/comprehensiveness (i.e. as post-hoc methods), suggesting that we need more robust metrics to evaluate post-hoc explanation faithfulness.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑