Refined Commonsense Knowledge from Large-Scale Web Contents

The pipeline is executed in the following order:

1. nlp_pipeline.pipeline
2. open_ie.open_ie
3. triple_filtering.filter
4. triple_grouping.group_per_c4_part, triple_grouping.group_all, triple_grouping.get_frequent_triples
5. triple_clustering.precompute_embeddings, triple_clustering.clustering
6. conceptnet_mapping.inference
7. ranking
8. final_filtering.final_filtering

Global configurations can be found in app_config.py.

Files needed for the pipeline to run are:

• Precomputed similarity scores between C4 documents and Wikipedia articles: https://nextcloud.mpi-inf.mpg.de/index.php/s/nJSSW5QBQR3XoxH (cf. triple_filtering/filter.py)
• Subjects: https://nextcloud.mpi-inf.mpg.de/index.php/s/TiSm3rrJ9kEqfm8 (cf. triple_filtering/filter.py)
• ConceptNet mapping train/dev files: https://nextcloud.mpi-inf.mpg.de/index.php/s/JeLRgsiNykcnRbs (cf. conceptnet_mapping/finetune.py)

If you use Ascent++, please cite the following paper:

@ARTICLE{ascentpp,
  author={Nguyen, Tuan-Phong and Razniewski, Simon and Romero, Julien and Weikum, Gerhard},
  journal={IEEE Transactions on Knowledge and Data Engineering}, 
  title={Refined Commonsense Knowledge from Large-Scale Web Contents}, 
  year={2022},
  doi={10.1109/TKDE.2022.3206505}
}