usage

1. Install the Required Packages

   • python version: 3.8.10
   • pip install -r requirement.txt

2. Convert ATOMIC 2020 Triples to Natural Language

   • download ATOMIC 2020 from https://allenai.org/data/atomic-2020 and put it in the directory "kb_process"
   • cd ./kb_process
   • python atomic_process.py
   • "atomic.csv" will be generated in the directory "datasets"

3. Convert ConceptNet 5.7 Triples to Natural Language

   • download conceptnet 5.7 from https://s3.amazonaws.com/conceptnet/downloads/2019/edges/conceptnet-assertions-5.7.0.csv.gz and put it in the directory "kb_process/concept_process"
   • cd ./kb_process/concept_process
   • python extract_cpnet_relation.py
   • python conceptnet-process.py
   • "conceptnet.csv" will be generated in the directory "datasets"

4. Transferring Knowledge from Large NLI Datasets

   • download MNLI from https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FMNLI.zip?alt=media&token=50329ea1-e339-40e2-809c-10c40afff3ce
   • download QNLI from https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FQNLI.zip?alt=media&token=c24cad61-f2df-4f04-9ab6-aa576fa829d0
   • download roberta-base from https://huggingface.co/roberta-base
   • download roberta-large from https://huggingface.co/roberta-large
   • put the datasets and models in the directory "/home/YOUR_USER_NAME/.cache/"
   • sh run_nli.sh /home/YOUR_USER_NAME/.cache/
   • models roberta+QNLI/MNLI will be stored in the directory "/home/YOUR_USER_NAME/.cache/"

5. Extract Knowledge from KBs

   • run the jupyter notes in the directory "kb_extract"
   • set the value of ROOT_DIR to "/home/YOUR_USER_NAME/NLI-KB/"
   • set the value of CACHE_DIR to "/home/YOUR_USER_NAME/.cache/"

6. Start Experiments

   • run the jupyter notes in the directory "roberta_unsup"
   • set the value of ROOT_DIR to "/home/YOUR_USER_NAME/NLI-KB/"
   • set the value of CACHE_DIR to "/home/YOUR_USER_NAME/.cache/"