comparator

Simple tool to compare almost similar names which are coming from the same source (for example list of all company owners and officers of that company). Helps to cluster together persons with a slight difference in name spelling/typos. Better suited for Cyrillic names, but should work everywhere.

It’s heuristic based and as any algorithm of such a nature it make errors sometime (see Accuracy section below for details).

Installation

Install from PyPI.

$ pip install names_comparator

Accuracy

You can test your installation on a ground truth data and check confusion matrix by running:

$ python comparator/__init__.py comparator/data/ground_truth.csv
    +--------------------+----------+----------+
    |                    | Positive | Negative |
    +--------------------+----------+----------+
    | Predicted positive |   290    |    9     |
    | Predicted negative |    14    |   687    |
    +--------------------+----------+----------+
    Precision:  0.97
    Recall:  0.95
    F1 score:  0.96

You can also run it with debug flag to see all the errors algorithm made:

$ python comparator/__init__.py comparator/data/ground_truth.csv yes

Usage

>>> from comparator import full_compare
>>> full_compare("Barack Hussein Obama", "Obama, Barak")
True
>>> full_compare("Петро Мазепа", "Мазепа Петро")
True
>>> full_compare("Марченко Петро Миколайович", "Панченко Петро Миколайович")
False
>>> full_compare("Овдієнко Сергій Костантинович", "Овдієнко Сергій Костянтинович")
True
>>> full_compare("Іванов Михайло Юрійович", "Іванов Юрій Михайлович")
False