SGPT: A Generative Approach for SPARQL Query Generation from Natural Language Questions

PyTorch code for the IEEE Access paper: SGPT: A Generative Approach for SPARQL Query Generation from Natural Language Questions [PDF].

⚙️ Installation (anaconda)

conda create -n sgpt -y python=3.8 && source activate sgpt
pip install -r requirements.txt
python -m spacy download en_core_web_sm

🏋️ Training

In multiple GPU setting, run the following command:

python -m torch.distributed.launch train.py --dataset lcquad2 --epochs 40

For single GPU:

python train.py --dataset lcquad2 --epochs 40

• valid dataset names: lcquad2, qald9, vquanda
• For using the masked entity in the question use --masked for both training and evaluation.
• For taking the additional knowledge (entities) into account --knowledge. Only entities are regarded as additional knowledge.

🎯 Evaluation

python -u eval.py --generate runs/sgpt/lcquad2/ --dataset lcquad2 --generation_params_file config/gpt-2-base/generation_params.json --eval_dataset test  --output_file outputs/predictions_gpt2-base.json

🎲 Hyper-paramters

Please try the following number of epochs to find the best results: 10,20,30,40 or 70 and the following learning rates: 6e-4 or 6e-5 .

📝 Citation

If you use the code, please cite the following paper.

@ARTICLE{
      9815253,  
      author={Rony, Md Rashad Al Hasan and Kumar, Uttam and Teucher, Roman and Kovriguina, Liubov and Lehmann, Jens},
      journal={IEEE Access},   
      title={SGPT: A Generative Approach for SPARQL Query Generation From Natural Language Questions},   
      year={2022},  
      volume={10},  
      number={},  
      pages={70712-70723},  
      doi={10.1109/ACCESS.2022.3188714}
    }

📜 License

MIT

📪 Contact

For further information, contact the corresponding author Md Rashad Al Hasan Rony (email).