Managing Data Labels

Create labeling session

1. In the Data Labeling page, select Choose from OOB labeling template.
   
   

   

   
   
2. In the Configure data labeling page, select the out-of-the-box package you want to configure. For the purpose of this example, we will use Email. The Email template is a combination of text classification and entity recognition labels.
3. Enter a name for the Data Labeling section and choose the Dataset. You can also create a new dataset from this page. For more information on datasets, see the About Datasets page.
4. Click Create.

Configure

In the Configure tab, you can configure the labels needed for your session.

There are two ways in which you can configure the labels:

• Using a General UI
• Using the Advanced editor

The left panel is the label config editor, while the right panel shows a preview of the current config. The right panel also shows the text components that are a part of the template. In this example, the Email template comes preset with a From, To, CC, Date and Text fields. You can add/remove these text components using the Advanced Editor.

Note: The text components configured here determines the structure of the files imported for labeling.

Using the interface

You can add new labels by using the interface from the General tab.

To add a label, click on ➕ next to Entity Label. Add the following information for the new label:

• Name: the name of the new label.
• Input to be labelled: choose which text component you want to be labeled. In the case of Email AI, you chose between Date, From, To, cc, and text.
• Attribute name: choose label names for the entities in your data. Use this to create attributes such as customer name, city name, telephone number, and so on.
• Shortcut: choose a hotkey for the label. Use this to configure a key as a shortcut. Only single letters or digits are supported.
• Color: choose a color for the label to distinguish it from the others.

Note: You can add more attributes for the same label by clicking on + Add new.

To add a classification, click on ➕ next to Classification. Add the following information for the new label:

• Name: the name of the new classification.
• Classification Type: choose the type of classification. You can choose between Single and Multiple.
• Attribute name: choose label names for each class or value in your data. Use this to create attributes such as sentiment (positive or negative), the type of email (customer, marketing, spam), and so on.
• Shortcut: choose a hotkey for the attribute. Use this to configure a key as a shortcut. Only single letters or digits are supported.

Note: You can add more attributes for the same classification by clicking on + Add new.

Using the Advanced Editor

You can add new labels by using the Advanced Editor tab.

To add a new label, make sure to use the following format: <Label value="NewLabel" background="color"/>. Add the labels in the <Labels name> section. Once you add the new label, it will be displayed on the right side of the screen.

Once you finish adding all the needed labels, click Done.

Note: Once configured, you can also reconfigure the labels by clicking Reconfigure.

Note that if you want to reconfigure data labeling, all labeled data that has not been exported will be lost and needs to be sent to labelers again.

Import

In the Import tab you can import the needed files for labeling. The labeling tool supports files in the JSON format. The JSON file should contain a data object that in turn contains the structure configured in the previous step. Here's a sample JSON file structured for the default Email template, containing From, To, CC, Date and Text fields as name-value pairs within a 'data' object.

{
  "data": {
    "date": "1/29/2021 12:29:01 PM",
    "from": "[e-mail]",
    "to": "[e-mail]",
    "cc": "",
    "text": "I currently have 4 separate student loans that I have been paying on for years with Citi,..."
  }
}

Once added, click Upload.

You can check all your uploaded files in the Uploading and Validating Files window. If there is an error to any of the files, the error message will help you in editing the file in order to meet the requirements.

For example, the cc field is missing in one of the uploaded files.

Once you checked the uploaded and validated files, click Done.

Labels

Note:

• If you are using Chinese or Japanese, granularity="word" must be removed using the Advanced Editor.
• If you are using right-to-left languages, you can set view style using the Advanced Editor. To change the alignment, add view style="text-align:right" in the Advanced Editor.

AI Center

Once you imported the files, you can send them to UiPath® Action Center for human in the loop and labeling. There are several ways to do this from AI Center. For more information, see the scenarios below.

A selection of files

To send only a selection of files to Action Center, follow the steps below:

1. Select the files you want to send from the list by marking the checkbox from their respective row.
2. Click Go to Action Center.
3. In the Send files to Action Center window, pick the folder from Action Center where you want to send the files to.
4. Check the Current selection bullet.
5. Click Ok.
   
   

   

   
   

All files

To send all files to Action Center, follow the steps below:

1. Click Go to Action Center.
2. In the Send files to Action Center window, pick the folder from Action Center where you want to send the files to.
3. Check the All bullet.
4. Click Ok.
   
   

   

   
   

Files not pushed

To send only the files that were not already sent to Action Center or labeled, follow the steps below:

1. Click Go to Action Center.
2. In the Send files to Action Center window, pick the folder from Action Center where you want to send the files to.
3. Check the All not pushed bullet.
4. Click Ok.
   
   

   

   
   

   Note: You can delete an existing label from the list, by clicking ⁝ next to the desired label and selecting Delete.

Action Center

Once the files are sent to Action Center, you can start labeling them.

1. Go to Action Center.
2. The files sent from AI Center are listed in My Actions under the Unassigned tab with the title Data labelling task.
3. Assign the task to yourself or someone else.
4. Go to the Pending tab and label the file accordingly.
   Note: The Annotations section is optional.
5. Click on ⁝ in the Data labelling task item from the list and select Edit labels to edit the labels.
   
   

   

   
   
6. Click on ⁝ in the Data labelling task item from the list and select Edit properties to edit the task.
   
   

   

   
   
7. Once the task is done, click Complete.
   
   

   

   
   Once the task is completed, you can check the status of the file in the Labels tab from the Data Labeling page in AI Center. To do so, click Refresh.The status should now be updated to Labeled and the number of Labeled items is changed.

Export

You can export labels from the Export tab.

To export files, click on Export files to dataset in the Export tab.

The export generates JSON files containing an 'annotations' object, in addition to the original 'data' object.

After the export is completed successfully, you can find the files at the following path:

Datasets > Dataset_name > exportedFiles

Note: If you do not provide a name in the Export Name field, the export files will have a timestamp for title.