Extract Key Phrases with AI

1. Understanding the Key Phrase Extraction Model

Like the text recognition model we used previously, key phrase extraction is a prebuilt model that requires no training. It takes a string of text as input and returns a list of the most important words and phrases.

To get a precise understanding of its capabilities and limitations, let's consult the official documentation.

From the documentation, here are the essential takeaways for a developer:

• Function: It extracts key talking points from unstructured text.
• Input: A text string, up to 5,120 characters.
• Output: A list of strings, where each string is an identified key phrase.

Now, let's put this into practice by building a Canvas App.

2. Implementing Key Phrase Extraction in a Canvas App

We will build a simple app that allows a user to type in text, click a button, and see a list of extracted key phrases. This process involves adding the AI model to the app, writing a Power Fx formula to call it, and binding the results to a control for display.

The following guide explains the general pattern for using any AI model with Power Fx in Power Apps. While its example uses the "Language detection" model, the principles and steps are identical for our "Key phrase extraction" model.

Step-by-Step Implementation

Let's apply what you've learned from the resources.

1. Create a Blank Canvas App:

   • Go to make.powerapps.com.
   • Create a new, blank Canvas App (tablet or phone layout).

2. Add the AI Model:

   • In the app editor, go to the Data tab on the left pane.
   • Click Add data, search for "AI models."
   • Select Key Phrase Extraction from the list to add it to your app.

3. Design the User Interface:

   • From the Insert tab, add the following controls to your screen:
     • A Text input control. Rename it txtFeedbackInput. Set its Mode property to TextMode.MultiLine to allow for more text. Clear its Default text.
     • A Button control. Rename it btnAnalyzeText and set its Text property to "Analyze Text".
     • A Vertical gallery control. Rename it galKeyPhrases. This will display our list of results.

4. Write the Power Fx Formula:

   • As recommended in the best practices, we will call the model from the button's OnSelect event to efficiently manage API calls.
   • Select the btnAnalyzeText button.
   • Set its OnSelect property to the following formula:

     ClearCollect(
         colKeyPhrases,
         'Key phrase extraction'.Predict(txtFeedbackInput.Text).Phrases
     )
     

   • Formula Breakdown:
     • 'Key phrase extraction'.Predict(txtFeedbackInput.Text): This calls the AI model, passing the text from our input box.
     • .Phrases: As per the documentation, the model's output is a record containing a table named Phrases. We access this table.
     • ClearCollect(colKeyPhrases, ...): This function takes the resulting table of phrases and stores it in a local collection named colKeyPhrases. ClearCollect ensures the collection is cleared before new items are added, so you don't mix results from previous clicks.

5. Display the Results:

   • Select the galKeyPhrases gallery control.
   • Set its Items property to colKeyPhrases.
   • Inside the gallery, select the default Title label.
   • Set its Text property to ThisItem.Value. ThisItem.Value refers to the text of the key phrase in the current row of the collection.

Now, run your app! Enter some text into the input box (e.g., "The hotel had a fantastic view and the staff was very friendly, but the Wi-Fi was unreliable.") and click the "Analyze Text" button. The gallery should populate with the key phrases: "fantastic view," "staff," "friendly," and "unreliable Wi-Fi."

Conclusion

Congratulations on completing the "Applying AI Builder Models" module! In this lesson, you successfully integrated the key phrase extraction model directly into a Canvas App. This enables you to build more intelligent and responsive user interfaces that can understand and process unstructured text in real time.

Key Takeaways:

• Model Integration: AI Builder models are added to Canvas Apps as data sources, making them available to use in Power Fx formulas.
• Power Fx Syntax: You call the model using the .Predict() function, passing the required inputs (like a text string). The syntax is consistent across models: 'Model name'.Predict(input).
• Handling Output: The key phrase model returns a table of phrases, which is best handled by storing it in a collection (ClearCollect) and displaying it in a gallery.
• Event-Driven Calls: Triggering the AI model on a button's OnSelect event is more efficient than using an input's OnChange event, which helps manage AI Builder credit consumption.

What's Next?

With your knowledge of integrating AI into both Power Automate and Power Apps, you have a solid foundation in one of the most powerful aspects of the Power Platform.

Our next module, "Power Apps Component Framework (PCF) Development," will transition into pro-developer territory. This is a topic you expressed specific interest in, and it's a critical skill for the PL-400 exam. In the first lesson, we will dive right into the practical setup, starting with installing the Power Platform CLI and initializing your first PCF component project. This will allow you to build fully custom, code-based controls that extend the capabilities of Model-driven and Canvas Apps far beyond what is possible out-of-the-box.