Skip to content

Test Queries

You can test your entity extractor model interactively by typing or copying and pasting pieces of text and examining any detected segments. Alternatively, you can retrieve a random document from a dataset associated with the model. While testing, assess whether the model's prediction response meets your standards. If it doesn't, make updates to the Duet model, teach, and test again.

  1. Press the Test tab at the top nav bar to start testing your model interactively.

  2. Type a query into the text box and click "Test".

  3. The model prediction will be shown inline on the query as underlines for each entity and sub-entity detected in the text.

    Additionally, the count of each detected entity type will show on the right. The count will show the number of detected segments for each schema entity including the parent entity. The system detected a single address with its city, state and zip components but not street, so more teaching would be required.

  4. You can view the model prediction results in JSON, and switch back and forth between the JSON and list view. The JSON view will show the predictions in the same way you will see it when you publish your model and send live traffic to it. The JSON view shows each detected entity and provides its type, value in the text, its position in the document as a character start index and its length (number of characters from the start index). If the prediction is a parent entity with children, they will be broken apart in sub-entities. You can see that "Address" has "City" and "State" as nested sub-entities.

  5. If you want to edit your labeling decisions or the schema, press "Resume Training" in the top-right. Otherwise, press "Publish".

Back to top