Annotation Tool


2OS AI Builder provides an end-to-end solution to create, annotate, manage, iterate, and train a model. Those advanced features allow the users to accelerate the creation of their models in a very fast and simple way with the Annotation Tool.

To create a new Machine Learning model for an AI application, you need to feed the engine with data enriched with meta-information. This process is generally called Annotation or Tagging.

Annotating a document consists of highlighting key elements in text or image content using Tags. A set of tags used during an annotation process is called a tagset. The tagged elements are then converted into data and used to train an AI model for future predictions of elements of the same type in other documents.


  • Tags and labels: The Annotation Tool allows the users to create, add and delete tags and labels (sub information within a tag) in a simple way, and makes the creation of a tagset very simple. The color palette allows you to assign specific colors to tags and to color code them.

  • Leveling up your annotations: At the beginning of the process, if the users choose to annotate all their data themselves, the annotation process can take a long time. This is why when creating your model, you can start by annotating data, then use the pre-annotation module to speed up the process.

  • Importing annotations from already existing datasets: The Annotation Tool allows the users to use annotations from external sources easily.

  • Two ways to select information depending on the final objective: The Annotation Tool can be set up on Bounding box mode. This mode allows you to draw bounding boxes around elements you want to highlight, for example, the outlines and cells of a table or around a figure.

  • The default mode of the Annotation Tool allows the users to select and highlight textual information, whether it is in a table or figure or not.

How it works

To create a new Machine Learning model, the user can prepare its own personalised data by annotating documents directly with the Annotation Tool. These annotations will then be used via the trained model to predict the same kind of information previously tagged on other documents. 

Figure 1. 2OS Annotation Tool

Figure 2. Title annotation example

Figure 3. Image annotation example

Process of a training dataset preparation: 

  • Step 1 : Choose your annotation mode ‘Bounding box‘ or ‘Text field
  • Step 2 : Create your tagset
  • Step 3 : Download the data to annotate
  • Step 4 : Start annotating
  • Step 5 : Save
  • Step 6 : 👉 Train your model using 20S AI modules 


  • Easy to use: An intuitive interface that allows quick annotation and labeling flexibility.

  • Possibility to improve your model by iteration: The ability to test your model and return to it by iteration.

  • Different formats available: The possibility to easily import your model and data in different formats.

  • Use of quality indicators: Quality indicators calculated automatically to help you improve your annotations.

Use Case

Classical NER: Person, Organization, Job, Location, Date.

Anaphora resolution: Labeled spans can be linked to each other by a relation. This functionality allows the user to create datasets on coreference. 

Sentiment annotation: positive and negative spans, emotions, opinions, recommendations.

Table extraction: In the Annotation Tool, two types of annotations are possible, depending on the final objective. If annotations on images or tables are needed, the Annotation Tool must be set up on Bounding box mode. This mode allows you to draw bounding boxes around elements you want to highlight, for example, tables and cells, in order to detect and extract the information contained in those tables.

Title annotations: This means annotating with a tag what you consider to be a title.

=======================TO SKIP=========================

(The objective of the 2OS platform is to give users the ability to create customized tools to work on their own data. We provide self-service models and a solution to create custom models.

Our first goal with 2OS is to give users the ability to use, discover, prepare and create their own application and also to help non-experts users to be more involved in technical details.

2OS will make Artificial Intelligence directly and easily accessible to non-experts users, saving time and resources as well as opening you up to an endless number of possibilities.)