Manually processing more documents in the training module in a consistent way improves your AI model's overall accuracy.
You can improve your Workflow AI model's accuracy by adding more documents to the training module and manually categorizing those documents, then selecting and labelling the content you're interested in extracting. Follow the steps below.
1. To add more documents to the Training module:
- Choose the Workflow you want to improve by clicking the
Workflows tab on the upper left, then select the desired Workflow. - Click the Menu button
on the top right corner, then click Edit Workflow.
2. To improve your Workflow AI model's Assisted Categorization function:
- Click the Categorize step of your Workflow, click the Vertical Ellipsis
button of the desired category, and click Modify. - Drag and drop your documents into the drag and drop space. We recommend you conduct each training set with at least 10 sample documents.
- After categorizing your sample documents, click the Close button on the bottom right. You can see the Overall Accuracy score's status (low, medium, high) shown below the Training Progress section.
Tip:
Click the Test Categorization button on the right and add some sample documents to check if the AI categorization performance has improved. You can also see the Overall Accuracy score's status (low, medium, high) shown below the Training Progress section.
3. To improve your Workflow AI model's Assisted Labelling function:
- Click the Label step of your Workflow.
- Select the document category in which you have added new documents and would like to improve the labelling feature by clicking the Document Category drop-down on the top right of the labelling screen and make your selection.
- Begin manually labelling the content of interest and remain consistent. Always select and label content in each document within the selected category the same way.
- After labelling a few documents, you will notice the assisted labelling function becoming activated. Content in the following documents will be automatically selected and labelled. Your task in this stage is to confirm or correct the AI's selections. The AI's confidence in its predictions is represented by a color-coating:
(Red = Low, Orange = Medium, Green = High). - Review each document's AI-generated labels and ensure that the AI has created all the desired labels. Then click each of the AI-generated labels and Delete or Confirm the selection.
- After you've finished training your Workflow AI model's Assisted Labelling function, click the Next Step button on the bottom right, which directs you to the Review step of your Workflow.
- If necessary, make any adjustments in the Review step, then click the Finish button on the bottom right.
Summary
Your Workflow's custom AI models improve by adding more documents to the training module, manually categorizing those documents, and labelling the content consistently within each document category.
Each document within a given category should have the desired content labelled identically, e.g., a client address should always be labelled "client address" on all documents.
Do not label a client address as "client address" on one document and label the same content as "customer address" on another document.