binonymizer

Binonymizer is a tool in Python that aims at tagging personal data¹ in a parallel corpus.

For example, for a input like:

URL1  URL2  My name is Marta and my email is fake@email.com    Mi nombre es Marta y mi email es fake@email.com

Binonymizer's output will be:

URL1 URL2 My name is <entity class="PER">Marta</entity> and my email is <entity class="EMAIL">fake@email.com</entity> Mi nombre es <entity class="PER">Marta</entity> y mi email es <entity class="EMAIL">fake@email.com</entity>

Detectable entity tipes

Currently, the Binonymizer is able to detect and tag the following types of entities:

• PER: person names
• ORG: organism and company names
• EMAIL: email addresses
• PHONE: phone numbers
• ADDRESS: addresses
• ID: personal card IDs (such as spanish DNIs)
• MISC: other personal data, or when the type it's uncertain
• OTHER: other

Installation & Requirements

Binonymizer works with Python 3.6, and can be installed with pip:

python3.6 -m pip install binonymizer

After installation, two binary files (binonymizer and binonymizer-lite) will be located in your python/installation/prefix/bin directory.

Language-dependant packages and models are automatically downloaded and installed on runtime, if needed.

Extra instructions for basque

In case you plan to binonymize basque data, you need to download binonymizer from github, and run the following steps:

cd binonymizer
git submodule sync
git submodule update --init --recursive --remote
cd prompsit_python_bindings
python3.6 setup.py install

Please note that you need to have access to Prompsit's private repository. Contact us if you need further details.

Usage

Binonymizer can be run with:

binonymizer [-h] --format {tmx,cols} [--tmp_dir TMP_DIR]
                     [-b BLOCK_SIZE] [-p PROCESSES] [-q] [--debug]
                     [--logfile LOGFILE] [-v]
                     input [output] srclang trglang

Parameters

• positional arguments:
  • input: File to be anonymized (See format below)
  • output: File with anonymization annotations (default: standard output)
  • srclang: Source language code of the input
  • trglang: Target language code of the input
• optional arguments:
  • -h, --help: show this help message and exit
• Mandatory:
  • --format {tmx,cols}: Input file format. Values: cols, tmx ("cols" format: URL1 URL2 SOURCE_SENTENCE TARGET_SENTENCE [extra columns] tab-separated)
• Optional:
  • --tmp_dir TMP_DIR: Temporary directory where creating the temporary files of this program (default: default system temp dir, defined by the environment variable TMPDIR in Unix)
  • -b BLOCK_SIZE, --block_size BLOCK_SIZE: Sentence pairs per block (default: 10000)
  • -p PROCESSES, --processes PROCESSES: Number of processes to use (default: all CPUs minus one)
• Logging:
  • -q, --quiet: Silent logging mode (default: False)
  • --debug: Debug logging mode (default: False)
  • --logfile LOGFILE: Store log to a file (default: standard error output)
  • -v, --version: show version of this script and exit

Example

binonymizer corpus.en-es.raw corpus.en-es.anon en es --format cols  --tmp_dir /tmpdir -b50000 -p31 

This will read the corpus "corpus.en-es.raw", which is in a column-based format, extracting personal data and writing the tagged output in "corpus.en-es.anon". Binonymizer will run in blocks of 50000 sentences, using 31 cores, and writing temporary files in /tmpdir

Lite version

Although binonymizer makes use of parallelization by distributing workload to the available cores, some users might prefer to implement their own parallelization strategies. For that reason, a single-thread version of the script is provided: binonymizer_lite. The usage is exactly the same as for the full version, but omitting the blocksize (-b) and processes (-p) parameter.

TO DO

• Fully support TMX input/output
• Address recognition
• GPU support
• Automate Prompsit-python-bindings submodule ( git submodule update --remote , python3.6 setup.py install)

¹: See EC definition of "personal information": https://ec.europa.eu/info/law/law-topic/data-protection/reform/what-personal-data_en