Skip to main content

Training

Create high-quality training data by labeling task files and marking them for AI model training across different processing steps.

Training Data Workflow​

Mark for Training​

Flag correctly labeled tasks as training data:

  1. Complete Labeling - Apply all necessary labels to a task file
  2. Review Accuracy - Verify labels are correctly placed and complete
  3. Mark for Training - Click the "Mark for training" button (Alt+A)
  4. Training Dataset - Task becomes part of your AI model training data

Marked tasks display a "Marked for training" badge and contribute to model improvement when sufficient training data is available.

Manual Prediction​

Generate AI predictions on-demand for any task:

  1. Select Task - Open the task file in the editor
  2. Predict - Click "Predict" button or use Alt+P hotkey
  3. Review Results - Examine AI-generated predictions
  4. Validate - Correct any inaccurate predictions manually
  5. Training Mark - Mark corrected predictions for training data

Erase All Labels​

Clear all labels and predictions to start fresh:

  1. Select Task - Open the task file requiring label removal
  2. Clear Labels - Click "Erase" button or use Alt+C hotkey
  3. Confirm Removal - All manual and predicted labels are removed
  4. Relabel - Apply new labels manually or trigger prediction

Step-Specific Labeling​

Different processing steps require different labeling approaches and support unique interaction methods.

Classification Labels​

Categorize documents and define routing logic:

Label Application​

  • Keyboard Shortcuts - Assign Ctrl + key combinations for quick labeling
  • Multiple Categories - Apply multiple classification labels per task
  • Confidence Thresholds - Set minimum prediction confidence levels

Right-Click Removal​

Remove classification labels using context menus:

  1. Right-click on applied classification label
  2. Select "Remove" from context menu
  3. Label removed instantly without confirmation

Hotkey Configuration​

Set up keyboard shortcuts for efficient labeling:

  • Ctrl Key - Toggle Ctrl key requirement for shortcut
  • Suffix Key - Define the letter/number for the combination
  • Quick Access - Label documents rapidly with keyboard shortcuts

Segmentation Labels​

Define document regions and content boundaries:

Visual Segmentation​

Create segments directly on document pages:

  • Manual Drawing - Draw segment boundaries on document image
  • Coordinate-Based - Define precise segment coordinates
  • Visual Feedback - Segments display with colored boundaries

Right-Click Editing​

Modify or remove segments using context menus:

  1. Right-click on existing segment
  2. Select action from context menu:
    • Edit - Modify segment boundaries
    • Delete - Remove segment completely
  3. Visual Update - Changes reflected immediately on document

Automatic Segmentation​

AI-assisted segment creation with manual review:

  • Model Predictions - AI suggests segment boundaries
  • Manual Adjustment - Refine predicted segments as needed
  • Approval Process - Review and approve segments before processing

Extraction Labels​

Annotate specific data fields and content areas:

Entity Placement​

Place extraction labels on target content:

  • Text Selection - Select text to annotate with entity labels
  • Field Mapping - Connect selected text to extraction fields
  • Multi-Selection - Apply same label to multiple text selections

Right-Click Management​

Remove or modify extraction annotations:

  1. Right-click on placed annotation
  2. Context Options:
    • Edit Label - Change annotation type or properties
    • Remove - Delete annotation completely
    • Copy - Duplicate annotation settings
  3. Immediate Update - Changes applied without additional confirmation

Field Descriptions​

Enhance AI model understanding with descriptive labels:

  • Label Names - Clear, descriptive field identifiers
  • Descriptions - Detailed explanations helping AI identify correct content
  • Examples - Sample text patterns the label should match

Grouping Labels​

Organize document pages into logical groups:

Context Grouping​

Define page relationships and grouping rules:

  • Page Selection - Choose pages belonging to same group
  • Group Names - Assign descriptive names to page groups
  • Relationship Rules - Define how pages relate within groups

Right-Click Operations​

Manage page groups efficiently:

  1. Right-click on grouped pages
  2. Group Options:
    • Ungroup - Remove pages from current group
    • Rename Group - Change group identifier
    • Add Pages - Include additional pages in group
  3. Visual Organization - Group changes reflected in page layout

Label Quality Guidelines​

Consistency Standards​

Maintain high-quality training data:

  • Consistent Naming - Use standardized label names across similar documents
  • Complete Coverage - Label all relevant content, not just obvious examples
  • Accurate Boundaries - Ensure precise text selection and segment boundaries
  • Field Validation - Verify extracted content matches expected data types

Training Data Balance​

Build robust datasets across document varieties:

  • Document Types - Include examples from all document categories
  • Content Variations - Label different layouts, formats, and content styles
  • Edge Cases - Include challenging examples and unusual formats
  • Sufficient Volume - Accumulate adequate examples for reliable training

Continuous Improvement​

Iteratively refine training data quality:

  • Review Predictions - Regularly examine AI predictions for accuracy
  • Correct Errors - Fix inaccurate predictions and mark corrected versions
  • Update Labels - Refine label definitions based on prediction performance
  • Monitor Performance - Track prediction accuracy over time

Keyboard Shortcuts​

Universal Actions​

Available across all step types:

  • Alt + A - Mark/unmark task for training
  • Alt + P - Trigger prediction on current task
  • Alt + C - Clear all labels from current task

Classification Shortcuts​

Step-specific keyboard combinations:

  • Ctrl + [Key] - Apply assigned classification label
  • Custom Keys - Configure in classification label settings

Troubleshooting​

Training Mark Issues​

Common problems and solutions:

  • Button Disabled - Ensure task has required labels and processing is complete
  • Missing Badge - Verify task status and refresh page if necessary
  • Unmarking Failed - Check task permissions and processing status

Prediction Problems​

Address prediction accuracy issues:

  • No Predictions - Verify model is configured and has sufficient training data
  • Inaccurate Results - Review and correct predictions, then mark for training
  • Model Errors - Check label mapping between project and model configurations

Labeling Interface​

Resolve labeling interface problems:

  • Right-Click Not Working - Ensure proper browser settings and page loading
  • Shortcuts Disabled - Verify keyboard shortcuts in browser and system settings
  • Visual Issues - Refresh page or check browser compatibility

Related Documentation​