Note
You must be registered as a manager for the project.
Select New Project in the Global Navigation bar.
Enter a name for the project. Choose a name that will be meaningful to you and your annotation team.
Select a Template type from the drop down menu. The templates are predefined and customized for each type of project.
Add annotators (and adjudicators) to the project. Click on Add/remove annotators to select from the users in the system. Users must be added to the system before you can add them to a project.
Configure the project. Each template type has its own set of configuration options.
Create.
Your new project will appear in the Project List.
Each template type has its own set of configuration options. In many cases the defaults will be acceptable. Here we list the options most likely to require adjustment.
All Templates:
Model Language: The language of the samples.
Hide Eval Predictions: By default, the Active Learning server offers suggested annotations for each sample. Suppress these on the samples in the evaluation set to avoid biasing the human annotators. Setting this option tells the Studio to only display suggestions for samples in the training set, and not for those in the evaluation set.
Prioritize Partially Annotated Docs: Prioritizes the samples presented for annotation such that full documents are completely annotated.
Show Token Boundaries: This option only affects the presentation of the text samples. When turned on, each unannotated span is underlined to make it easier to see the token spans.
Mouse Full Token Selection: When you click on or select part of the text, the entire token is selected.
Part-of-Speech (POS):
Events:
Maximum Number of Candidate Clusters: (default=10) The maximum number of candidates under consideration during the role extraction pass.
Event Similarity Threshold: (default 0.3) Minimum score for the system to extract an event.
Text Span Similarity Threshold: (default 0.4) Minimum similarity of embedding scores on TEXT chunks.
Entity Text Bias: (default 1.2) Bias factor for matches between ENTITY and TEXT chunks.
Most project configuration options can be changed after the project is created. Click on the project menu in the upper right corner of the project and select Configure. Some options, such as model language, cannot be changed once the project has been created.
Note
You must be registered as a manager for the project.
Select Add Document in the action bar at the bottom of the project’s dashboard.
(Optional) Give the document a name. Some documents are named automatically; for example, each file in an ingested zip file will create a document named with the filename.
-
Add one or more documents.
(Optional) Assign annotators. By default, each added document will be given to all the annotators currently assigned to the project. If you add annotators to the project later, they will not be automatically be given documents that were added earlier.
-
Add the documents into the project and select whether the document(s) will be used for training or validation.
Add for training: documents used to train the model
Ad for validation: golden data used to test the model's accuracy
Add (auto-split): automatically assign documents to the training and validation sets
Note
Documents that are identical will be merged. For pure-text documents it automatically deduplicates documents. For .adm documents, which may include prior annotations or metadata, the additional data is merged into a single document.