Privacy Enhancement for Text via Risk-oriented Explainability (PETRE)

This repository contains the code and data for the text anonymization enhancement method presented in B. Manzanares-Salor, D. Sánchez, Enhancing text anonymization via re-identification risk-based explainability, Submitted, (2023). In addition, some results obtained during testing are provided.

The data was extracted from the bootstrapping-anonymization repository, corresponding to the publication Papadopoulou, A., Lison, P., Øvrelid, L., Pil ́an, I., 2022. Bootstrapping text anonymization models with distant supervision, in: Proceedings of the Thirteenth Language Resources and Evaluation Conference, European Language Resources Association, Marseille, France. pp. 4477–4487. It has been extended by adding boidies of individuals' biographies and annotations made by multiple automatic anonymization approaches. The resulting data can be found in the Datasets/Wiki553 folder.

The code is presented in the PETRE.ipynb notebook. The project can be run locally (using Jupyter) or in Google Colab. If Colab is used, it is necessary to upload the contents of this repository to a Google Drive folder, so that Colab can access it. Use the Text Re-Identification repository, corresponding to the publication B. Manzanares-Salor, D. Sánchez, P. Lison, Automatic Evaluation of Disclosure Risks of Text Anonymization Methods, Privacy in Statistical Databases, (2022), for obtaining the required re-identification pipeline.

Project structure

Privacy Enhancement for Text via Risk-oriented Explainability
│   README.md                                       # This README
│   PETRE.ipynb                                     # Code notebook
└───Datasets                                        # Folder for all datasets
│   └───Wiki553                                     # Dataset folder
│       └───Annotations                             # Folder for all anonymizations
│       │   │   Annotations_Manual.json             # List of manually-crafted annotations
│       │   │   Annotations_Presidio.json           # List of annotations made by Presidio
│       │   │   Annotations_Gold.json               # Human-based reference annotations for precision/recall evaluation
│       │   │   ...
│       │
│       └───Wiki553.json                            # Pandas dataframe with abstracts and bodies of individuals' biographies
└───Outputs                                         # Folder for PETRE's results for all datasets
    └───Wiki553                                     # Results for the used dataset
        └───BG_Abstracts                            # Results only using articles' abstracts as background knowledge
        │   └───Annotations_St.NER4                 # Folder for results using Stanford NER4 as starting anonymization
        │   │   │   Annotations_PETRE_Start.json    # Initial annotations after PETRE's chunking
        │   │   │   Annotations_PETRE_k=2.json      # PETRE annotations for k=2
        │   │   │   ...
        │   │    
        |   └───Original_Ranks                      # Ranks per document for each starting anonymization
        │   │   │   ...
        │   │
        │   └───PETRE_Ranks                         # Ranks per document for each PETRE's output
        │       │   ...
        │
        └───BG_Bodies                               # Results only using articles' bodies as background knowledge
        │    │   ...
        │
        └───BG_Bodies+Abstracts                     # Results using both articles' bodies and abstracts as background knowledge
            │   ...