Find in Library

Named Entity Recognition (NER) poses challenges for both flat and nested tasks, which require different paradigms. To overcome this issue, we present GFNER, a unified global feature-aware framework based on table filling, that can handle both types of tasks with low computational cost. While pretrained models have shown great promise in NER, they typically focus on local contextual information, disregarding global relationships that are crucial for accurate entity boundary extraction. To address this limitation, we introduce a global feature learning module that captures the inter-entity associations and significantly enhances entity recognition. Experimental results on flat and nested NER datasets demonstrate that GFNER outperforms previous state-of-the-art models. The code for GFNER is available at <uri>https://github.com/cjymz886/GFNER</uri>.