HierGAT

This is the implementation of "Entity Resolution via Hierarchical Graph Attention Network"

Environment

• Python 3.7
• PyTorch 1.4
• HuggingFace Transformers
• NLTK (for 1-N ER problem)

You should run pip install -r requirements.txt first.

Datasets

The raw datasets can be found at

• Magellan: https://github.com/anhaidgroup/deepmatcher/blob/master/Datasets.md
• WDC: http://webdatacommons.org/largescaleproductcorpus/v2/index.html
• DI2KG: http://di2kg.inf.uniroma3.it/datasets.html

Train HierGAT

python train.py \ 
	--task Amazon \
	--batch_size 32 \
	--max_len 256 \
	--lr 1e-5 \
	--n_epochs 10 \
	--finetuning \
	--split \
	--lm bert

• --task: the name of the tasks (see task.json)
• --batch_size, --max_len, --lr, --n_epochs: the batch size, max sequence length, learning rate, and the number of epochs
• --split: whether to split the attribute, should always be turned on
• --finetuning: whether to finetune the LM, should always be turned on
• --lm: the language model. We now support bert, distilbert, xlnet, roberta (bert by default)
  • If you want to load the model file locally, you can configure the --lm_path

Train HierGAT+

python train_n.py \ 
	--task N/Amazon \
	--su_len 10 \
	--finetuning \
	--split \
	--lm bert

Same as HierGAT, with one additional parameter:

• --su_len: max entity-level context sequence length