Efficient Nearest Neighbor Promptbased Learning for Fewshot Ner in Manufacturing

The NER task in manufacturing is usually lack sufficient labeled data resourcesTo tackle this issuethis paper presents an effective NNPLM framework for fewshot NER in manufacturingwhich introduce a simple enhancement of the promptbased learning model using nearest neighbor retrieval We retrieve the morphologically similar characters for each character to be predicted and then rectifies the prediction Moreoverwe use supervised contrastive learningSCLand instance weighting to get better semantic representations of multicategory charactersCompared with the best baseline our NNPLM achieves a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{7. 1 2 \%}$</tex> F1 score average improvement on all fewshot settings in manufacturing