Research map / Label-efficient and robust learning / Recognition and long-tail
Weakly Supervised Sequence Tagging from Noisy Rules
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 linking rules and linked HMM instead of HMM without linking rules (bi-LSTM trained) (ablation) | bi-LSTM-CRF NER · NCBI-Disease | F1: 71.35 -> 79.03 | Helped |
| add linking rules and linked HMM instead of HMM without linking rules (bi-LSTM trained) (ablation) | bi-LSTM-CRF NER · LaptopReview | F1: 62.86 -> 69.04 | Helped |
| add linking rules and linked HMM instead of HMM without linking rules (bi-LSTM trained) (ablation) | bi-LSTM-CRF NER · BC5CDR | F1: 82.21 -> 82.96 | Mixed |
| use linked HMM instead of Snorkel naive Bayes for denoising rules (ablation) | bi-LSTM-CRF NER · NCBI-Disease | F1: 73.41 -> 79.03 | Helped |
| use linked HMM instead of Snorkel naive Bayes, direct predictions (ablation) | generative model · NCBI-Disease | F1: 68.72 -> 78.66 | Helped |
| use linked HMM instead of HMM, direct predictions (ablation) | generative model · BC5CDR | F1: 83.01 -> 85.87 | Helped |
| model rule accuracy: unweighted vote vs linked HMM | bi-LSTM-CRF NER · BC5CDR | F1: 82.43 -> 82.96 | No clear effect |
| use linked HMM instead of majority vote (ablation) | bi-LSTM-CRF NER · LaptopReview | F1: 62.51 -> 69.04 | Helped |
| use linked HMM weak supervision for SRL instead of supervised baseline on small dev set (ablation) | bi-LSTM-CRF SRL · Ontonotes SRL | F1: 63.06 -> 69.00 | Helped |
| use linked HMM instead of unweighted vote (ablation) | bi-LSTM-CRF SRL · Ontonotes SRL | F1: 61.96 -> 69.00 | Helped |
| use linked HMM instead of naive Bayes (ablation) | bi-LSTM-CRF SRL · Ontonotes SRL | F1: 58.93 -> 69.00 | Helped |
| weak supervision (linked HMM) versus full supervision | bi-LSTM-CRF NER · NCBI-Disease | F1: 87.16 -> 79.03 | Hurt |
Closest papers by meaning
- Transfer Learning for Sequence Tagging with Hierarchical Recurrent Networks (ICLR 2017 · similarity 0.89)
- WRENCH: A Comprehensive Benchmark for Weak Supervision (NeurIPS 2021 · similarity 0.87)
- Low Resource Sequence Tagging with Weak Labels (AAAI 2020 · similarity 0.87)
- Learning from weak labelers as constraints (ICLR 2025 · similarity 0.87)
- Disambiguation of Weak Supervision leading to Exponential Convergence rates (ICML 2021 · similarity 0.86)
- Merging Weak and Active Supervision for Semantic Parsing (AAAI 2020 · similarity 0.86)
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.