SConE: Simplified Cone Embeddings with Symbolic Operators for Complex Logical Queries

This is the implementation of the paper SConE: Simplified Cone Embeddings with Symbolic Operators for Complex Logical Queries (Findings of ACL 2023).

Getting started

Step 1: Data preparation

• Download the datasets here, then move KG_data.zip to ./scone/ directory

• Unzip KG_data.zip to ./scone/data/:

  cd scone/
  unzip -d data KG_data.zip

Step 2: Installing requirements

• If you are familiar with pip|conda, please install requirements by your own preference:

  python=3.8.11
  pytorch=1.9.1
  numpy=1.19.2
  tqdm=4.65.0
  tensorboardX=2.5.1

• [Optional] For those prefer to use Anaconda, create virtual environment named scone (default) with dependencies then activate scone:

  conda env create -f requirements.yml
  conda activate scone

Step 3: Training model

• [Optional] run the following bash command to train model for the default dataset FB15k-237, uncomment others in run.sh to train model using other datasets (FB15k/NELL995):

  ./run.sh

• Otherwise, use the direct command in the following to train scone (FB15k-237, etc.).

FB15k-237

CUDA_VISIBLE_DEVICES=0 python main.py --cuda \
--data_path data/FB15k-237-betae \
--do_train --do_test \
-n 128 -b 512 -d 400 -g 20 \
-lr 0.00005 --max_steps 350001 --cpu_num 2 --geo scone --valid_steps 30000 \
-projm "(1600,2)" --save_checkpoint_steps 30000 -logic "geometry" \
--seed 0 --print_on_screen -p 0.9 -projn "rtrans_mlp" -conj "all" -delta 0.5

FB15k & NELL995 (see run.sh)

Citation

If you find this code useful for your research, please consider citing the following paper:

@inproceedings{nguyen2023scone,
    title = "{SC}on{E}: Simplified Cone Embeddings with Symbolic Operators for Complex Logical Queries",
    author = "Nguyen, Chau D. M. and
      French, Tim  and
      Liu, Wei  and
      Stewart, Michael",
    booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.findings-acl.755/",
    doi = "10.18653/v1/2023.findings-acl.755",
    pages = "11931--11946",
}

Acknowledgement

We acknowledge the code of KGReasoning for their contributions.