ner_concept.md

Potential Research Concepts for the NER Task

Architecture

• arch-rnn: Recurrent Neural Networks (LSTMs, GRU)
• arch-cnn: Convolutional Neural Networks
• arch-transformer: Transformer
• arch-gnn: Graph Neural Networks (GNN)
• arch-att: Attention Mechasnim
• arch-crf: Conditional Random Field
• arch-semicrf: Semi-Conditional Random Field

Training

• train-multitask: Multi-task learning
• train-multimodal: Multi-modal learning
• train-auxiliary: Joint training
• train-transfer: Cross-domain learning, Transfer or domain adaptation
• train-multiling: Bi-lingual, Multi-lingual learning
• train-active: Active learning, Bootstrapping
• train-adver: Adversarial learning

Pre-trained Models

• pre-train: Pre-trained, Contextualized (BERT)

Task Setting

• task-nested: Nested or Multi-grained NER
• task-hetero: Heterogeneous categories
• task-lowres: Low-resource or Zero-resource learning
• task-weaksup: Weakly Supervised NER
• task-disc: Discontinuous NER

Dataset

• dataset: Constructing a new dataset

Interpretable Analysis

• interpret-genz: Generalization
• interpret-model: System-wise Analysis
• interpret-error: Annotation Errors Analysis