Research map / Label-efficient and robust learning / Domain adaptation and generalization
CrossNER: Evaluating Cross-Domain Named Entity Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add span-level masking rather than token-level masking during integrated-corpus DAPT (ablation) | BERT pre-train then fine-tune · CrossNER average | F1: 68.29 -> 69.63 | Helped |
| Use the integrated entity-level and task-level corpus instead of entity-level alone (ablation) | BERT pre-train then fine-tune with token-level masking · CrossNER average | F1: 67.24 -> 68.29 | Helped |
| Use the integrated entity-level and task-level corpus instead of task-level alone (ablation) | BERT pre-train then fine-tune with token-level masking · CrossNER average | F1: 66.71 -> 68.29 | Helped |
| Add DAPT (ablation) | BERT directly fine-tuned on target domains · CrossNER average | F1: 61.44 -> 66.55 | Helped |
| Add DAPT (ablation) | BERT pre-trained on source then fine-tuned on target · CrossNER average | F1: 64.89 -> 69.63 | Helped |
| Add DAPT (ablation) | BERT jointly trained on source and target · CrossNER average | F1: 64.52 -> 68.72 | Helped |
| Pre-train on source NER data before target fine-tuning instead of fine-tuning only on target data (ablation) | BERT without DAPT · CrossNER average | F1: 61.44 -> 64.89 | Helped |
| Jointly train on source and target instead of fine-tuning only on target (ablation) | BERT without DAPT · CrossNER average | F1: 61.44 -> 64.52 | Helped |
| Use task-level corpus instead of entity-level corpus (ablation) | BERT pre-train then fine-tune with token-level masking · CrossNER average | F1: 67.24 -> 66.71 | Hurt |
| Add span-level masking and integrated-corpus DAPT to the existing state-of-the-art model (ablation) | Existing state-of-the-art cross-domain NER model · CrossNER average | F1: 66.55 -> 68.71 | Helped |
| Use source-domain training data in specialized entity categories (ablation) | CrossNER Music entities · Music test: band and musical artist | F1 | Hurt |
Closest papers by meaning
- A Simple Yet Effective Subsequence-Enhanced Approach for Cross-Domain NER (AAAI 2023 · similarity 0.95)
- Three Heads Are Better than One: Improving Cross-Domain NER with Progressive Decomposed Network (AAAI 2024 · similarity 0.93)
- Domain-Adapted Dependency Parsing for Cross-Domain Named Entity Recognition (AAAI 2023 · similarity 0.92)
- MCL-NER: Cross-Lingual Named Entity Recognition via Multi-View Contrastive Learning (AAAI 2024 · similarity 0.91)
- UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition (ICLR 2024 · similarity 0.88)
- ConsistNER: Towards Instructive NER Demonstrations for LLMs with the Consistency of Ontology and Context (AAAI 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.