This repository contains scripts/recipes for text manual annotation using Prodigy annotation tool by Explosion (the creators of spaCy). In order to use Prodigy for your annotation tasks you need a license, but researchers can apply for an interim license to use Prodigy free in their research.
The following recipes were inspired in code found in Explosion AI's GitHub repository with the help of Prodigy's Support Forum.
The focus should not be on the specific example applications I used but on the code to create the blocks of annotation tasks - that you can use in your own applications.
Recipe for manually annotating text:
- identify named entities (for named entity recognition);
- classify text into categories (for binary classification or multiclass classification into 3 classes).
In the example below, the annotatores can highlight 3 named entities in sentences: weather related keywords, location and date/time. The squared buttons at the bottom are used for the data classification task: to classify text into weather related, non-weather related or unsure.
Recipe for manually annotating text:
- identify named entities (for named entity recognition);
- classify text into categories (for binary classification or multiclass classification;
- free-text input field;
- checkbox to create a flag variable.
In the example below, the annotators can highlight opinion/sentiment expressions in reviews, classify each review as positive, negative, both or neutral, write more about their decision, point out if they think the user who wrote the review was being sarcastic and accept, reject or ignore the annotation in hand.
