sev777/cldr_clner_models

Though language model text embeddings have revolutionized NLP research, their ability to capture high-level semantic information, such as relations between entities in text, is limited. In this paper, we propose a novel contrastive learning framework that trains sentence embeddings to encode the relations in a graph structure. Given a sentence (uns

Imposing Relation Structure in Language-Model Embeddings Using Contrastive Learning

Official code repository of "Imposing Relation Structure in Language-Model Embeddings Using Contrastive Learning" paper (PyTorch implementation). For a description of the models and experiments, see our paper: https://aclanthology.org/2021.conll-1.27/ (published at CoNLL 2021).

Setup

Requirements

• Python 3.5+
• PyTorch (tested with version 1.3.1)
• scikit-learn (tested with version 0.24.2)
• tqdm (tested with version 4.55.1)
• numpy (tested with version 1.17.4)
• rapids (tested with version 0.19)
• seaborn (tested with version 0.11.2)

Execution steps

• Download the CharacterBERT model [1] using the official repository.
• Run the data_preprocessing.py script to process the data ade_full_spert.json.
• Run the offline_graph_creation.py script to extract the graphs for each sentence.
• Train the CLDR and CLNER models by running the main.py script under the corresponding folder. There are two modes, the tuning mode, and the final run. If you want to execute only the final run, consult the final_run_epochs.xlsx file to find the number of training epochs per split (cross-validation).
• Run the embeddings_RE_NER_jointly.py to extract the trained embeddings.
• Solve the NER and RE tasks by running the KNN classifiers under the /classification/ folder. The "final run" mode is implemented.
• Run the evaluation.py script to extract the evaluation metrics per task. Strict evaluation [2] is used.
• For the tSNE [3] analysis, first run the dataset_creation.py script to extract the dataset in a particular format (.hdf5 file). Then run the tSNE_*.py scripts to create the plots.
  

*** For each execution step the corresponding (if any) config file (/configs/) should be updated accordingly. Importantly, change the split_id number.

Notes

Please cite our work when using this software.

Theodoropoulos, C., Henderson, J., Coman, A. C., & Moens, M. F. (2021, November). Imposing Relation Structure in Language-Model Embeddings Using Contrastive Learning. In Proceedings of the 25th Conference on Computational Natural Language Learning (pp. 337-348).

BibTex:

@inproceedings{theodoropoulos2021imposing,
  title={Imposing Relation Structure in Language-Model Embeddings Using Contrastive Learning},
  author={Theodoropoulos, Christos and Henderson, James and Coman, Andrei Catalin and Moens, Marie Francine},
  booktitle={Proceedings of the 25th Conference on Computational Natural Language Learning},
  pages={337--348},
  year={2021}
}

References

[1] Hicham El Boukkouri, et al. 2020. "Characterbert: Reconcilingelmo and bert for word-level open-vocabulary rep-resentations  from  characters." In Proceedings of the 28th International Conference on Computational Linguistics, pages 6903–6915.
[2] Bruno Taillé, et al. 2020. "Let’s stop error propagation in the end-to-end relation extraction literature!" In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP 2020), pages 3689–3701.
[3] Laurens  Van  der  Maaten  and  Geoffrey  Hinton.  2008. "Visualizing  data  using  t-sne." Journal  of  MachineLearning Research, 9(11).