Home

Welcome to the documentation of the atr-ner-eval library!

It provides an easy way to compute a wide variety of metrics to evaluate automatic workflows for Automatic Text Recognition (ATR) and Named Entity Recognition (NER).

Use case

The main application of this toolbox is the evaluation of ATR-NER models.

From this document image

The following text and entities are predicted automatically (with mistakes):

    You see , it was , opparently , through a mistake on
    Guy
    's pert that we missed seeing the flemingoes our
    first
    morning on the island . "
    " What harm could possibly save come to
    Forrest through ) Sir
    John
    's nonsense ? "
    Piens
    could barely have spoken with more contempt .
    A bully like that respects anyone who ceres to stand up to him .. "

But the actual text and entities are as follows:

    You see , it was , apparently , through a mistake on
    Guy
    's part that we missed seeing the flamingoes our
    first
    morning
    on the island . " " What harm could possibly have come to
    Forrest
    through Sir
    John
    's nonsense ? "
    Piers
    could hardly have spoken with more contempt .
    A bully like that respects anyone who dares to stand up to him . "

This library gives insights to easily evaluate the performance of the automatic workflow with respects to a set of manually annotated examples.

Get started

Get started with the library here.