DocsICP FilingConsole
Documentation
Home Page AgentLoop User Guide Data Center Data annotation

Data annotation

Updated at:
Copy as MD
Product
My Favorites

The data annotation feature lets you add structured labels to records in a dataset. It supports multiple annotation widget types and helps annotators complete annotation tasks efficiently. This topic describes the end-to-end annotation workflow.

Step 1: Bind an annotation template

In the Data Center, select a dataset and navigate to the Annotation tab to create an annotation template.

  1. Open the details page of the target dataset.

  2. Switch to the Annotation tab.

  3. Click Create Annotation Template to open the template binding page.

Step 2: Configure the annotation task

Set the field sources, annotation widgets, and annotation instructions.

Field sources

Field sources define which data columns are displayed in the annotation workspace.

  • Add a field: Click + Add Field to add a display field.

  • Select a data column: Each field must be bound to a dataset column from the dropdown list. For JSON-type columns, the system automatically expands two levels of sub-paths for selection (for example, metadata.user.name).

  • Set the display type: Each field can be displayed as text, image, or JSON. The system automatically infers the default type based on the data format.

  • Add a field description: Add a description to each field to help annotators understand its meaning.

  • Reorder fields: Drag the handle to adjust the display order of fields.

  • Remove a field: Click the delete button next to a field to remove it.

Note

If no field sources are configured, the annotation interface displays all fields in the dataset by default. You can also click Select All Fields to bind all top-level fields at once.

Annotation widgets

Annotation widgets define the annotation controls on the right side of the annotation workspace. Click + Add Widget Type to add a widget. The following five widget types are supported:

Widget type

Name

Description

Text

TextArea

Free-form text input for open-ended annotations such as summaries, comments, or corrected text.

Choices

Choices

Single-select or multi-select from predefined options.

Rating

Rating

Numeric scoring with a configurable range (for example, 1 to 5). Suitable for quality scoring or satisfaction assessment.

Boolean

Boolean

Binary choice (Yes/No). Suitable for pass/fail or relevance judgments.

Taxonomy

Taxonomy

Hierarchical category labels with single-select or multi-select mode. Suitable for multi-level classification systems.

Each widget supports the following configurations:

  • Widget title: Set the title displayed above the widget to help annotators understand its purpose.

  • Option list (Choices and Taxonomy widgets): Add, edit, and delete options to define the available answers.

  • Score range (Rating widget): Set the minimum and maximum values for the scoring range.

  • Multi-select mode (Taxonomy widget): Toggle between single-select and multi-select mode by selecting the Multi-select checkbox.

Note

You can add multiple widget types to a single annotation task for multi-dimensional annotation. For example, combine a Rating widget for quality scoring with a TextArea widget for notes.

Annotation instructions

Enter annotation guidelines or standard requirements in the Annotation Instructions area at the bottom of the configuration section. This helps annotators understand the task requirements and annotation criteria.

Preview and save

  • Right-side preview: All configuration changes are reflected in the preview area in real time. Use the preview to verify the final layout of the annotation interface.

  • Save configuration: Click Start Annotating to save the annotation configuration to the current dataset.

Step 3: Start annotating

After saving the configuration, click Start Annotating to enter the annotation workspace and annotate records one by one.

Two ways to start annotating

Method

Description

Annotation tab

On the dataset details page, switch to the Annotation tab to enter the annotation workspace directly.

Data list button

On the dataset list page, click the annotation button on a data row to open the annotation workspace in a side panel.

Annotation workspace layout

The annotation workspace uses a two-panel layout:

Area

Position

Description

Data display

Left

Displays the bound field content for the current record, rendered according to the configured field sources and display types.

Annotation controls

Right

Displays all annotation widgets. Annotators complete their annotations in this area.

You can drag the divider between the two panels to adjust their width ratio, which is helpful when viewing long text or complex data.

The top of the data display area shows the current record ID and annotation status (annotated / not annotated).

Browse and filter data

The top of the annotation workspace provides data filtering and navigation features:

  • All / Not Annotated toggle: In the upper-right corner, switch between viewing all data or only unannotated data. The record count is displayed next to each option.

  • Progress bar: The top area shows the annotation progress bar and the number of annotated records.

  • Previous / Next: Use the buttons at the bottom to navigate sequentially between records.

  • Skip: Click Skip to temporarily skip the current record and continue with the next one.

Annotation history

For previously annotated records:

  • The system automatically loads the previous annotation results and populates the widgets.

  • Click Annotation History to toggle the display of historical annotation data.

  • If the annotation template has changed and is incompatible with previous annotations, the system shows a "Template has changed" message and cannot load the old data.

Step 4: Annotate records

Annotation workflow

  1. Review the current record content in the left panel.

  2. Complete annotations using the widgets in the right panel:

    • TextArea: Enter annotation text in the text box.

    • Choices: Select one or more predefined options.

    • Rating: Click a star or number to assign a score.

    • Boolean: Select Yes or No.

    • Taxonomy: Select one or more category labels from the hierarchical category tree.

  3. After verifying the annotations, click the submit button to save the results.

  4. The system automatically navigates to the next unannotated record.

Submit and complete

  • After submission, the annotation results are saved to the current data record.

  • After submitting the last record, the system displays a completion summary showing the total number of records and the number of annotated records.

Reconfigure the template

In the annotation workspace, click the Rebind button next to the template name at the top to switch to a different annotation template.

Best practices

Annotation task design tips

Tip

Description

Keep fields minimal

Bind only the data columns needed for annotation to avoid information overload that reduces annotation efficiency.

Write clear descriptions

Add clear descriptions to each field to reduce the cognitive load on annotators.

Match widgets to use cases

Select widget types that match the annotation goal. Prefer selection-based widgets to improve annotation consistency.

Provide thorough instructions

Write complete annotation instructions that include criteria, edge cases, and examples.

Preview before annotating

Use the right-side preview to confirm the configuration before starting formal annotation.

FAQ

Issue

Solution

Annotation history cannot be loaded

The template may have changed and become incompatible. Re-annotate the records using the current template.

"Please complete annotation first" error on submit

Ensure you have interacted with at least one annotation widget before submitting.

Dataset has no data

Add data on the dataset details page before starting annotation.
Previous：Manage datasets in the consoleNext：AI annotation