A Neighborhood Framework for Resource-Lean Content Flagging

Abstract We propose a novel framework for cross- lingual content flagging with limited target- language data, which significantly outperforms prior work in terms of predictive performance. The is based on nearest-neighbor architecture. It modern instantiation the vanilla k-nearest neighbor model, as we use Transformer representations all its components. Our can adapt to new source- instances, without need be retrained from scratch. Unlike neighborhood-based approaches, encode neighborhood information query– interactions. two encoding schemes and show their effectiveness using both qualitative quantitative analysis. evaluation results eight languages different datasets abusive detection sizable improvements up 9.5 F1 points absolute (for Italian) over strong baselines. On average, achieve 3.6 improvement three Jigsaw Multilingual dataset 2.14 WUL dataset.