Education-ner-dataset

EduNER is a Chinese named entity recognition dataset for education research.

├── models
│   ├── BERT-CRF
│   ├── BERT-NER
│   ├── BiLSTM-CRF
│   ├── CLNER
│   ├── Flat-Lattice-Transformer
│   ├── Flert
│   ├── LEBERT
│   ├── LexiconAugmentedNER
│   ├── LGN
│   ├── LR-CNN
│   ├── MECT4CNER
│   ├── SLK-NER
│   └── TENER
├── Cohen_Kappa
├── comparison_Dataset
├── dataset
├── imgs

EduNER

• models/ directory contains the sampling version of our dataset.

• Quality: Cohen's Kappa consistency examination

• The related resource paper ✨ can be found in Neural Computing & Applications journal.

• A snapshot of entity types

• Reference

  • Li, X., Wei, C., Jiang, Z. et al. EduNER: a Chinese named entity recognition dataset for education research. Neural Comput & Applic (2023). https://doi.org/10.1007/s00521-023-08635-5

Models

basic

• models/ directory contains the recent SOTA models.
• Lexicon Augemented NER includes SoftLexicon+CNN/Transformer/LSTM models.
• CLNER includes the CL-KL and CL-L₂ models.

tutorial

• Pre-trained embedding

  We use the Chinese pre-trained character or word embeddings, e.g., ctb.50d, gigaword_chn.all.a2b.bi.ite50, and gigaword_chn.all.a2b.uni.ite50 in line with (Yang et al., 2017). We use pre-trained language model, the Chinese BERT:bert-base-chinese.

• Hyper parameters

  models	epoch	batch size	max length	learning rate	dropout rate
  example	100	10	256	0.001	0.5
  BiLSTM+CRF	100	32	Adaptive length	0.001	0.5
  BERT	20	32	256	5e-5	0.5
  BERT+CRF	20	16	256	3e-5	0.1
  LR-CNN	150	10	256	1.5e-3	0.5
  TENER	100	16	Adaptive length	7e-4	0.15
  LGN	10	1	256	2e-4	0.5
  FLAT+BERT	100	10	200	6e-4	0.5
  SoftLexicon (CNN)	100	30	256	5e-3	0.5
  SoftLexicon (Transformer)	100	30	256	5e-3	0.5
  SoftLexicon (LSTM)	100	30	256	5e-3	0.5
  MECT4CNER	100	10	200	1.4e-3	0.2
  SLK-NER	30	32	256	5e-5	0.5
  LEBERT	20	4	256	1e-5	0.1
  FLERT	10	4	512	5e-6	0.1
  CL-KL	10	1	512	5e-6	0.1
  CL-L2	10	2	512	5e-6	0.1

• Code instruction, reproduce benchmark models

Online Annotation Platform

• We provide a temporary account to test the annotation tool

username: edu
password: 

Update plan

To a long-term plan, EduNER dataset project, we expect the dataset to cover more languages and disciplines in higher education. Although this goal is not achieved in a short duration, the dataset will expand to one or two disciplines and will acquire a bigger scale dataset that can be used for teaching or learning contexts.

• Pedagogic Psychology discipline will be added in the future.
• Policy, Conference related corpus will be added in the future.

Beta application

• A beta educational tool ( EDUNERScore ) based on our dataset can be accessed. The tool is based on NER technology and allows for the analysis of unstructured educational texts in real-time. Specifically, the tool can extract the discipline entity from large-scale unstructured texts, e.g., discourse content, online forums, writing documents etc. It will help the stakeholder to better understand the learning or teaching activity.
• Due to limited computing resources, only cached results can be viewed at the current. In addition, only the Chinese version is now available.
• Instruction

License

Shield:

This work is licensed under a Creative Commons Attribution 4.0 International License.