Trainable splitter (Preview)

创建新项目

When creating a new project, tenants located in Europe, the US and Japan may enable our new splitter and classifier model. This trainable model can automatically split and classify complex documents, enabling you to turn messy packets into clean, typed documents.

Follow the instructions from this page to a Document Understanding^TM project and enable the new splitter and classifier model:

1. 打开 Document Understanding。
2. 选择“创建项目” 。
3. 填写所需的项目名称。
4. 选择“新式”以享受新式体验。
   Note: This is a guided model building experience that also includes recommendations for optimal model performance and active learning.
5. Switch on the Enable new splitter and classifier model toggle.
6. 如果需要，请配置高级选项。
   1. Switch on the Enable splitting toggle to enable the model to automatically split documents into individual files before classification. You can also enable this feature from the Project settings screen.
      Important: When the Enable splitting option is turned off, all documents are classified as a whole.
   2. 从“OCR 方法”下拉列表中选择要用于新项目的 OCR。
   3. 填写 OCR API 密钥。
      注意：如果您选择 UiPath™ OCR，则系统会自动填充此字段。
   4. 填写 OCR URL。有关 UiPath OCR URL 的完整列表，请查看“公共端点”页面。
   5. 选择是否要在 PDF 上应用 OCR。默认情况下设置为“自动”。
7. 选择“创建”。
   Figure 1. Figure 1. Creating your first project

成功创建项目后，您可以从“构建”部分上传文档。

Choose one of the two available options:

• Extract data from documents: pulls specific fields, such as invoice numbers, dates, total from your documents etc.
  Note: We recommend choosing this option when you need structured data for automation or reporting.
• Classify and split documents: sorts documents by type and separates multiple documents within a single file.
  Note: We recommend choosing this option when you need to organize and prepare documents for extraction.

Extracting data from documents

1. Select a document type.
2. Select Upload or drag and drop your files inside the new document type. Wait for the upload to finish.

Classify and split documents

Certain complex files contain several document types. Our new model can detect where each sub-document starts and ends and classify each section accordingly.

1. Click on Classify and Split Documents and upload your document packets. Wait for the document to finish uploading and processing.

2. Select any documents from the upload section and click Split. This will open up the splitting annotation interface.

   
   Note: If the project already has a trained model, uploaded documents will be pre-annotated using that model. This helps speed up annotation and allows you to view prediction results on new documents.

3. Click New document type to create a document type for each item in your desired taxonomy. You can select a predefined document type or create a custom one.

   
   Note: For custom document types, provide a name, a short description explaining its purpose, and comma-separated key indicators (such as unique fields or terms) that help identify it.

4. Indicate where documents should be separated. Assign each page range to a document type using the dropdown menu. Once you have finished annotating the document, click Confirm.

   
   Note: Clicking Confirm triggers document processing. After processing, each sub-document will appear under its corresponding document type in the Build section.
   Note: Each sub-document moved to a document type will get pre-annotated with the schema of the document type.

模型训练

Model training is triggered only after:

• At least five sub-documents have been created and annotated;
• A document has been confirmed.

The training status can be viewed in the upper right-hand corner of the Classification pane.

重要提示：

• The maximum document size is 160MB or 500 pages.
• Pages cannot be reordered or deleted.

Splitting and classification predictions

Whenever a new model is trained, all documents within the project receive predictions from the trained model. This allows you to review the performance of the classification model.

The “Type” column displays the ground truth, which is the document type as it was annotated. The “Predicted type” column shows the type predicted by the model.

By default, only the document packets are displayed in the UI. To view the sub-documents within each packet, click View and select the Include sub-documents checkbox.

Predictions can also be viewed in the annotation interface by enabling the “Show Prediction” toggle.