Doccano

Doccano is a self-hosted text annotation tool for machine learning practitioners who need labeled training or evaluation data. It runs on your own machine or server, with a browser interface and Docker support. The software is open source under the MIT license.

Its annotation tasks cover text classification, sequence labeling, and sequence-to-sequence work. You can assign sentiment labels, mark named entities within text, or pair a document with a summary. These task types let teams prepare datasets for different language models within the same application.

Projects support collaborative annotation, with multiple users and shared annotation guidelines. Doccano can import existing labeled datasets as well as data awaiting annotation, then export the results for use elsewhere. It handles multiple data formats and supports multilingual text, emoji, and access from mobile devices. The interface also has a dark theme.

For teams that already have scripts or models involved in labeling, the REST API connects Doccano to that work. A machine learning model can label data through the API, alongside the annotation interface used by people. This makes it relevant to both manual labeling projects and workflows that incorporate model-generated labels.

The application has a Python backend built with Django and Django REST Framework, plus a Vue.js and Nuxt.js frontend. Teams can customize it to their needs. It uses SQLite by default and also supports PostgreSQL.