ARAIDA: Analogical Reasoning-Augmented Interactive Data Annotation

Human annotation is a time-consuming task that requires a significant amountof effort. To address this issue, interactive data annotation utilizes anannotation model to provide suggestions for humans to approve or correct.However, annotation models trained with limited labeled data are prone togenerating incorrect suggestions, leading to extra human correction effort. Totackle this challenge, we propose Araida, an analogical reasoning-basedapproach that enhances automatic annotation accuracy in the interactive dataannotation setting and reduces the need for human corrections. Araida involvesan error-aware integration strategy that dynamically coordinates an annotationmodel and a k-nearest neighbors (KNN) model, giving more importance to KNN'spredictions when predictions from the annotation model are deemed inaccurate.Empirical studies demonstrate that Araida is adaptable to different annotationtasks and models. On average, it reduces human correction labor by 11.02compared to vanilla interactive data annotation methods.