Active learning for Relation Extraction and Assertion Status Models in the Annotation Lab

A new generation of the NLP Lab is now available: the Generative AI Lab. Check details here https://www.johnsnowlabs.com/nlp-lab/

We are pleased to announce the release of Annotation Lab version 2.2.

This version increases the number of models available on the Models Hub page for preannotation, supports the uploading of an image file (or multiple images zipped) from your computer. It also implements lots of improvements for the Visual NER Project and Labeling Page, as well as UI enhancements and fixed known issues. It can be used for such sophisticated fields as NLP in finance market or healthcare.

Relation Extraction and Assertion Status Models

Users can easily use Relation Extraction and Assertion Status Models for preannotation jobs. A valid Spark NLP for HealthCare License is needed to download them from the Models Hub page. After that as with other models, users can directly use them in the Project Config and deploy a model server.

Relation Extraction

Download and load relation extraction model

Preannotate using downloaded relation extraction model

Assertion Status Model

Preannotating Assertion statuses are also exactly the same as it is done for other project types. First, an admin user needs to download models from Models Hub and then Project Owner or Manager should configure Assertion Labels using the following two attributes:

1. add assertion=”true” attribute to indicate any Label as Assertion Status
2. add another attribute to associate the name of the model with the Label using model=”assertion_model_name”

Once it is deployed along with NER model(s), hit the preannotate button on one or more tasks from the Task List page.

With the addition of these two types, now Annotation Labs support all of the following types of models for preannotation.

1. Named Entity Recognition Models (Free/Paid)
2. Classification Models (Free/Paid)
3. Assertion Status Models (Paid Only)
4. Relation Extraction Models (Paid Only)
5. Embeddings (Free/Paid)

Local Images Upload

Until this version, only images from remote URLs could be uploaded for Image projects. With this version, the Annotation Lab supports uploading images from your local storage/computer. It is possible to either import one image or multiple images by zipping them together. It is possible for the upload to fail due to the file size being limited to 16 MB. If you need to upload files exceeding the default configuration, please contact your system administrator who will change the limit size in the installation artifact and run the upgrade script.

Visual NER

Our previous version of Annotation Lab introduced Visual NER Labeling Project. The current version improved functionality around this feature. A sample task can be imported from the Import page by clicking the “Add Sample Task” button. Also, the default config for the Visual NER project contains a zoom feature that supports the maximum possible width for low-resolution images when zooming.

Improved Relation Labeling

Creating numerous relations in a single task can look a bit clumsy. The limited space in the Labeling screen, the relation arrows, and different relation types all at once could create difficulty to visualize them properly. We improved the UX for this feature:

1. Spaces between two lines if relations are present
2. Ability to Filter by certain relations
3. When hovered on one relation, only that is focused

Get & install it here.

Full feature set here.