Research map / Architectures and efficiency / RNNs and normalization
Deep Active Learning for 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 |
|---|---|---|---|
| use CNN-CNN-LSTM instead of CNN-CNN-CRF (ablation) | CNN-CNN tag encoders · CoNLL-2003 English test | F1: 90.35 ± 0.24 -> 90.69 ± 0.19 | Helped |
| use CNN-CNN-LSTM instead of CNN-CNN-CRF (ablation) | CNN-CNN tag encoders · OntoNotes 5.0 English test | F1: 86.15 ± 0.08 -> 86.52 ± 0.25 | Helped |
| use an LSTM rather than CNN for the word-level encoder (ablation) | CNN-CNN-LSTM · CoNLL-2003 English test | F1: 90.69 ± 0.19 -> 90.58 ± 0.28 | Hurt |
| use an LSTM rather than CNN for the word-level encoder (ablation) | CNN-CNN-LSTM · OntoNotes 5.0 English test | F1: 86.52 ± 0.25 -> 86.40 ± 0.48 | No clear effect |
| use an LSTM rather than CNN for both character- and word-level encoders (ablation) | CNN-CNN-LSTM · OntoNotes 5.0 English test | F1: 86.52 ± 0.25 -> 86.63 ± 0.49 | No clear effect |
| increase the LSTM decoder beam size from 1 to 2 (ablation) | LSTM-LSTM-LSTM · OntoNotes 5.0 English test | F1: 87.26 -> 87.34 | Helped |
| increase the LSTM decoder beam size beyond 2 (ablation) | LSTM-LSTM-LSTM · OntoNotes 5.0 English test | F1: 87.34 -> 87.33 | No clear effect |
| select examples using active learning rather than random sampling | NER training · OntoNotes 5.0 English and Chinese | test F1 | Helped |
| use MNLP or BALD rather than least confidence in early acquisition rounds | NER active learning · OntoNotes 5.0 English | test F1 | Helped |
| normalize sequence log-probability by sentence length (MNLP) | least-confidence sentence selection · NER active learning | annotation efficiency | Helped |
| use representativeness-based sampling, alone or combined with uncertainty sampling | NER active learning · OntoNotes 5.0 | test F1 | Hurt |
| train on a random half of the data rather than genre-biased subsets | CNN-CNN-LSTM · OntoNotes 5.0 English | F1: 81.49 -> 85.10; 82.08 -> 85.10 | Helped |
Closest papers by meaning
- Small Language Model Makes an Effective Long Text Extractor (AAAI 2025 · similarity 0.86)
- Adversity-aware Few-shot Named Entity Recognition via Augmentation Learning (AAAI 2025 · similarity 0.86)
- UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition (ICLR 2024 · similarity 0.86)
- Span Graph Transformer for Document-Level Named Entity Recognition (AAAI 2024 · similarity 0.86)
- PaDeLLM-NER: Parallel Decoding in Large Language Models for Named Entity Recognition (NeurIPS 2024 · similarity 0.85)
- A Reasoning Paradigm for Named Entity Recognition (AAAI 2026 · similarity 0.85)
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.