EASAL: Entity-Aware Subsequence-Based Active Learning for Named Entity Recognition

Active learning is a critical technique for reducing labelling load by selecting the most informative data. Most previous works applied active on Named Entity Recognition (token-level task) similar to text classification (sentence-level task). They failed consider heterogeneity of uncertainty within each sentence and required access entire annotator when labelling. To overcome mentioned limitations, in this paper, we allow algorithm query subsequences sentences propose an Entity-Aware Subsequences-based Learning (EASAL) that utilizes effective Head-Tail pointer one entity-aware subsequence based BERT. For other tokens outside subsequence, randomly select 30% these be pseudo-labelled training together where model directly predicts their pseudo-labels. Experimental results both news biomedical datasets demonstrate effectiveness our proposed method. The code released at https://github.com/lylylylylyly/EASAL.