SGEL

Code for WWW 2020 paper: "High Quality Candidate Generation and Sequential Graph Attention Network for Entity Linking"

Dependencies

This project is based on python>=3.6. The dependent package for this project is listed as below:

tensorflow>=1.8.0
scikit-learn==0.21.3
xgboost==0.9

Training Command

1.Extract statistical features

python model/local_feature.py

2.Calculate xgboost score and filter candidate

python model/xgboost_rank.py

3.Get BERT embedding

python model/process_bert.py

Supplement: Due to historical factors, we train the local model based on the BERT source code. Now you can choose to use huggingface to train the BERT local model.

4.Rank mention

python model/local_ranker.py

5.Train the global GAT model

python model/selector.py