Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/bram-code/llm-anonymization

This repository provides utilities for anonymizing, pseudonymizing, and simplifying Dutch text using various NLP techniques.
https://github.com/bram-code/llm-anonymization

anonymization dutch large-language-models llm named-entity-recognition ner pseudonymisation simplification spacy

Last synced: 3 months ago
JSON representation

This repository provides utilities for anonymizing, pseudonymizing, and simplifying Dutch text using various NLP techniques.

Awesome Lists containing this project

README 

        
# LLM-anonymization



This repository provides utilities for anonymizing, pseudonymizing, and simplifying Dutch text using various NLP techniques. The main functionalities are implemented in the `anonymization_utils.py` file, with an example usage provided in the `example.ipynb` notebook.



## Features



- **Anonymization**: Replace sensitive entities in the text with generic placeholders.

- **Pseudonymization**: Replace sensitive entities with unique identifiers.

- **Text Simplification**: Simplify Dutch text using a pre-trained model.

- **SARI Score Calculation**: Evaluate the quality of text simplification.

- **Entity Extraction**: Extract named entities from the text.

- **Entity Comparison**: Compare entities between original and simplified texts.

- **Score Plotting**: Visualize the performance metrics.



## Usage



### Anonymization



To anonymize text, use the `anonymize_text` function:



```python

from anonymization_utils import anonymize_text



text = "Your text here"

anonymized_text, analyzer_results = anonymize_text(text)

print(anonymized_text)

```



### Pseudonymization



To pseudonymize text, use the `pseudonymize_text` function:



```python

from anonymization_utils import pseudonymize_text



pseudonymized_text = pseudonymize_text(text, analyzer_results)

print(pseudonymized_text)

```



### Text Simplification



To simplify text, use the `simplify_text` function:



```python

from anonymization_utils import simplify_text



simplified_text = simplify_text(anonymized_text)

print(simplified_text)

```



### SARI Score Calculation



To calculate the SARI score, use the `calculate_sari_score` function:



```python

from anonymization_utils import calculate_sari_score



sari_score = calculate_sari_score(text, simplified_text)

print(sari_score)

```



### Entity Extraction



To extract entities from text, use the `extract_entities` function:



```python

from anonymization_utils import extract_entities



entities = extract_entities(text)

print(entities)

```



### Entity Comparison



To compare entities between original and simplified texts, use the `compare_entities` function:



```python

from anonymization_utils import compare_entities



precision, recall, f1, true_positives, false_positives, false_negatives = compare_entities(original_entities, simplified_entities)

print(f"Precision: {precision}, Recall: {recall}, F1 Score: {f1}")

```



### Score Plotting



To plot the performance metrics, use the `plot_scores` function:



```python

from anonymization_utils import plot_scores



anonymized_scores = [precision_anonymized, recall_anonymized, f1_anonymized, sari_score_anonymized]

pseudonymized_scores = [precision_pseudonymized, recall_pseudonymized, f1_pseudonymized, sari_score_pseudonymized]



plot_scores(anonymized_scores, pseudonymized_scores)

```



## Evaluation Metrics



To ensure the effectiveness and accuracy of our text processing techniques, we use several evaluation metrics, particularly in the context of text simplification and anonymization/pseudonymization:



- **Precision**: Indicates the accuracy of the entities identified by the model. High precision means that the model correctly identifies and anonymizes sensitive information without affecting irrelevant data.

- **Recall**: Reflects the model's ability to find all relevant instances in the text. High recall ensures that most, if not all, sensitive information is identified and anonymized.

- **F1 Score**: Balances precision and recall, providing a single metric that considers both the accuracy and completeness of the anonymization process. A high F1 score indicates a well-performing model that effectively anonymizes sensitive information while minimizing errors.

- **SARI Score**: Evaluates the quality of text simplification by comparing the simplified text to the original and reference texts. It helps assess how well the simplified text retains its meaning and readability after anonymization.



These metrics complement each other by providing a comprehensive evaluation of the model's performance in both simplifying and anonymizing text. Precision and recall focus on the accuracy and completeness of anonymization, while the F1 score offers a balanced view. The SARI score ensures that the simplified text remains clear and understandable, even after anonymization.



## Example



For a complete example, refer to the `example.ipynb` notebook in this repository. It demonstrates how to use the utilities step-by-step with sample text.



## Conclusion



In this repository, we have provided a comprehensive set of tools for anonymizing, pseudonymizing, and simplifying Dutch text using advanced NLP techniques. By leveraging the functionalities in `anonymization_utils.py`, users can effectively protect sensitive information while maintaining the readability and integrity of the text. The example provided in `example.ipynb` demonstrates the practical application of these techniques, showcasing their effectiveness through various evaluation metrics.



These tools are particularly relevant in domains where privacy and data protection are paramount, such as healthcare, legal, and customer service industries. By ensuring that sensitive information is anonymized or pseudonymized, organizations can comply with data protection regulations while still utilizing valuable textual data for analysis and processing.



Overall, the combination of anonymization, pseudonymization, and text simplification techniques presented in this repository offers a robust solution for handling sensitive information in Dutch texts, making it a valuable resource for researchers and practitioners in the field of natural language processing and data privacy.



## Sources



Microsoft. (n.d.). Presidio Pseudonymization. Retrieved from https://microsoft.github.io/presidio/samples/python/pseudonymization/



Explosion AI. (n.d.). spaCy Documentation. Retrieved from https://spacy.io



Hugging Face. (n.d.). UL2-large-Dutch-Simplification-MAI-2023 Model. Retrieved from https://huggingface.co/BramVanroy/ul2-large-dutch-simplification-mai-2023



Xu, W., Napoles, C., Pavlick, E., Chen, Q., & Callison-Burch, C. (2016). Optimizing Statistical Machine Translation for Text Simplification. Transactions of the Association for Computational Linguistics, 4, 401-415. Retrieved from https://www.aclweb.org/anthology/Q16-1029



Explosion AI. (n.d.). spaCy Models for Dutch. Retrieved December 5, 2024, from https://spacy.io/models/nl



Sun, R., Jin, H., & Wan, X. (2021). Document-Level Text Simplification: Dataset, Criteria and Baseline. CoRR, abs/2110.05071. Retrieved from https://arxiv.org/abs/2110.05071



Project Jupyter. (n.d.). Retrieved December 5, 2024, from https://jupyter.org/



Matplotlib. (n.d.). Retrieved December 5, 2024, from https://matplotlib.org/



Espinosa-Zaragoza, I., Abreu-Salas, J., Lloret, E., Moreda, P., & Palomar, M. (2023). A Review of Research-based Automatic Text Simplification Tools. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2023). Association for Computational Linguistics. Retrieved from https://aclanthology.org/2023.ranlp-1.36.pdf



Aamot, H., Kohl, C. D., Richter, D., & Knaup-Gregori, P. (2013). Pseudonymization of patient identifiers for translational research. BMC Medical Informatics and Decision Making, 13, 1-15. Springer.



Marques, J. F., & Bernardino, J. (2020). Analysis of Data Anonymization Techniques. In KEOD (pp. 235-241).



Yermilov, O., Raheja, V., & Chernodub, A. (2023). Privacy-and utility-preserving NLP with anonymized data: A case study of pseudonymization. arXiv preprint arXiv:2306.05561.



Olatunji, I. E., Rauch, J., Katzensteiner, M., & Khosla, M. (2022). A review of anonymization for healthcare data. Big Data. Mary Ann Liebert, Inc.



Minh, P. Q. N. (2018). A feature-based model for nested named-entity recognition at VLSP-2018 ner evaluation campaign. arXiv preprint arXiv:1803.08463.