CANDLE 🕯: Extracting Cultural Commonsense Knowledge at Scale

Running spaCy on your input corpus

The first step is to run spaCy on your input corpus of choice. The script candle/run_spacy.py can be used for this purpose. For example, to run this script on the dummy files in the candle/data/input_corpus directory, run the following command:

cd candle
python run_spacy.py \
    -i data/input_corpus/dummy-000.jsonl \
    -o data/spacy/dummy-000.spacy

The input file should be a JSONL file, where each line is a JSON object with the following fields:

• text: The text of the document (required).
• timestamp: The timestamp of the document (optional).
• url: The URL of the document (optional).

After running spaCy on all the input files, you should create a file consisting of the paths to all the spaCy output files (see e.g., candle/data/spacy/dummy.txt). This file should be passed to the next steps using the spacy_file_list argument (see below).

CANDLE pipeline execution

There are 6 components (see candle/pipeline/pipeline.py):

1. candle/pipeline/component_people_group_matcher.py
2. candle/pipeline/component_generic_sentence_filter.py
3. candle/pipeline/component_culture_classifier.py
4. candle/pipeline/component_clustering.py
5. candle/pipeline/component_rep_generator.py
6. candle/pipeline/component_ranking.py

For example, to run the pipeline for the religions domain (see also candle/config_religions.yaml), follow these steps:

Start your local MongoDB instance:

cd /path/to/mongodb/folder
bin/mongod --dbpath /folder/to/save/the/database --bind_ip_all

Run the first 3 components:

cd candle/candle
python main.py \
  --config config_religions.yaml \
  --people_group religions \
  --spacy_file_list data/spacy/dummy.txt \
  --components 1 2 3

Run the last 3 components:

for facet in "food" "drink" "ritual"
do
  python main.py \
    --config config_religions.yaml \
    --people_group religions \
    --components 4 5 6 \
    --cluster_facet $facet \
    --cluster_nid data/religions/religion_ids.txt \
    --domain religions \
    --output_file _outputs/religions_$facet.jsonl
done

Citation

If you use this code or our datasets, please cite the following paper:

@inproceedings{candle2023,
  title={Extracting Cultural Commonsense Knowledge at Scale},
  author={Nguyen, Tuan-Phong and Razniewski, Simon and Varde, Aparna and Weikum, Gerhard},
  booktitle={Proceedings of the ACM Web Conference},
  year={2023}
}

More information is available on: https://candle.mpi-inf.mpg.de/