De-Bias for Generative Extraction in Unified NER Task

Zhang, Shuai; Shen, Yongliang; Tan, Zeqi; Wu, Yiquan; Lü, Weiming

doi:10.18653/v1/2022.acl-long.59

Cited by 20 publications

(17 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…First, it can be seen that our model achieves the best performance among all compared methods on almost all datasets. E.g., on GENIA, our model outperforms the sequence labeling [22] and span-based methods [13] by 5.07% and 4.67% in terms of F1 score, and increases over 1.57%, 0.54% and 0.69% compared with the seq2seq models [21,9,10]. The hypothesis is that our approach effectively models the entity relation, improving entity boundary detection for better entity generation.…”

Section: Resultsmentioning

confidence: 84%

“…We compare our model with three seq2seq models [21,9,10], and several other baselines that are specifically designed for individual NER substask, including sequence labeling [22], span-based methods [13] and hypergraph model [8], etc. The performance comparisons on three type datasets are reported 1, 2 and 3 respectively.…”

Section: Resultsmentioning

confidence: 99%

“…We evaluate flat NER datasets on the CoNLL-2003 following the same settings in previous works [5,15] and OntoNotes 5.0 which is adopted the splits as used in [9,15]. And we experiment on nested NER datasets GENIA following [9,13,10], ACE 2004, ACE 2005 in english following [9,15]. We conduct experiment on discontinuous NER datasets CADEC, ShARe13 and ShARe14 corpus and use [18] to process the data.…”

Section: Datasetsmentioning

confidence: 99%

“…α and τ in the focal loss of the entity relation are set to 5 and 1 respectively. Following the prior work [9,10], we compute the precision (P), recall (R) and F1 (F) scores and tune model according to the F1 on the development set.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Recently, sequence-to-sequence (seq2seq) generative approaches [9,10] have been proposed to jointly model all NER tasks in a unified framework, which attracts a lot of attention in the NER community. Although achieving promising results on all three NER categories, there are two major limitations of these generative models.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multi-task Transformer with Relation-attention and Type-attention for Named Entity Recognition

Ying¹,

Tang²,

Wang³

et al. 2023

Preprint

View full text Add to dashboard Cite

Named entity recognition (NER) is an important research problem in natural language processing. There are three types of NER tasks, including flat, nested and discontinuous entity recognition. Most previous sequential labeling models are task-specific, while recent years have witnessed the rising of generative models due to the advantage of unifying all NER tasks into the seq2seq model framework. Although achieving promising performance, our pilot studies demonstrate that existing generative models are ineffective at detecting entity boundaries and estimating entity types. This paper proposes a multi-task Transformer, which incorporates an entity boundary detection task into the named entity recognition task. More concretely, we achieve entity boundary detection by classifying the relations between tokens within the sentence. To improve the accuracy of entity-type mapping during decoding, we adopt an external knowledge base to calculate the prior entity-type distributions and then incorporate the information into the model via the self and cross-attention mechanisms. We perform experiments on an extensive set of NER benchmarks, including two flat, three nested, and three discontinuous NER datasets. Experimental results show that our approach considerably improves the generative NER model's performance.

show abstract