Research map / Continual learning / Online and replay-based
Continual Learning for Named Entity Recognition
Natawut Monaikul, Giuseppe Castellucci, Simone Filice, Oleg Rokhlenko
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 knowledge distillation from the previous model rather than training only on hard teacher labels (ablation) | ExtendNER · CoNLL-03 test, all steps and permutations | sentence-level F1: 0.37 points | Improved |
| Add knowledge distillation from the previous model rather than training only on hard teacher labels (ablation) | ExtendNER · OntoNotes test, all steps and permutations | sentence-level F1: 0.81 points | Improved |
| Extend the output layer instead of using separate output layers merged with heuristics (ablation) | NER continual learning · CoNLL-03 and OntoNotes | macro-F1: 0.06 points; 0.04 points | No clear effect |
| Initialize each update from the previous model rather than train from scratch on the current slice | ExtendNER · CoNLL-03 and OntoNotes test, all steps | sentence-level F1 | Improved |
| Use partially annotated new data rather than fully annotated data for old and new entity types | ExtendNER · CoNLL-03 test, all steps | sentence-level F1: 0.24 | Hurt |
| Use partially annotated new data rather than fully annotated data for old and new entity types | ExtendNER · OntoNotes test, all steps | sentence-level F1: 0.78 | Hurt |
| Allow encoder parameters to update while training on the new entity type, while keeping the old output layer frozen | Transfer learning for NER · CoNLL-03 test, step 2 | F1: 83.77 -> 34.95 | Hurt |
| Freeze the encoder while training the new entity output layer | Transfer learning for NER · CoNLL-03 test, step 2 | F1: 57.49 | Hurt |
| Add new entity types sequentially with ExtendNER and KD | ExtendNER · CoNLL-03 test, previously learned entity types | F1: +1.308 | Improved |
| Add new entity types sequentially with ExtendNER and KD | ExtendNER · OntoNotes test, previously learned entity types | F1: +0.0341 | No clear effect |
| Continue adding entity types with ExtendNER and KD | ExtendNER · OntoNotes test, e1 from step 1 to step 6 | F1: 78.03 -> 77.88 | No clear effect |
Closest papers by meaning
- Unify Named Entity Recognition Scenarios via Contrastive Real-Time Updating Prototype (AAAI 2024 · similarity 0.90)
- Prototypical Replay with Old-class Focusing Knowledge Distillation for Incremental Named Entity Recognition (AAAI 2025 · similarity 0.89)
- ConsistNER: Towards Instructive NER Demonstrations for LLMs with the Consistency of Ontology and Context (AAAI 2024 · similarity 0.89)
- UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition (ICLR 2024 · similarity 0.88)
- GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition (AAAI 2025 · similarity 0.88)
- Three Heads Are Better than One: Improving Cross-Domain NER with Progressive Decomposed Network (AAAI 2024 · similarity 0.87)
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.
