AI-102 Question 216
Select 3You are designing an Azure AI solution that uses a Language Understanding (LUIS) application to process customer support chatbot queries. The chatbot must identify intents such as 'CheckOrderStatus' and 'CancelOrder'. During testing, you notice the application is incorrectly identifying the 'CheckOrderStatus' intent for queries like 'I want to cancel my order'. Which of the following actions should you take to improve intent recognition accuracy?
- A
Provide more labeled utterances for each intent in the LUIS application.
- B
Increase the confidence score threshold for intent recognition in the application.
- C
Add a prebuilt entity to the LUIS application to identify key terms like 'order' and 'cancel'.
- D
Review and refine labeled utterances to ensure they are representative of user inputs.
- E
Enable active learning in LUIS to identify and retrain on misclassified utterances.
Show answer and explanation
Correct answers: A, D, E
Explanation
Improving intent recognition in a LUIS application involves providing more labeled utterances, refining the quality of utterances, and leveraging active learning to iteratively enhance the model. These methods directly improve the model's ability to accurately classify user queries by enhancing its training dataset and learning process. Adjusting thresholds or adding entities may address other issues, but they are not solutions to intent misclassification specifically.
- A. Correct.
Providing more labeled utterances for each intent helps the LUIS model better understand and distinguish between intents. It ensures the model has sufficient data to learn patterns for each intent.
- B. Incorrect.
Increasing the confidence score threshold does not directly improve the model's ability to distinguish intents; it merely changes the criteria for accepting an intent prediction.
- C. Incorrect.
Adding a prebuilt entity can help in extracting entities like 'order' or 'cancel', but it does not directly address the problem of incorrect intent recognition.
- D. Correct.
Refining labeled utterances ensures that they are diverse, realistic, and reflect actual user queries, which improves the model's ability to identify the correct intent.
- E. Correct.
Enabling active learning allows the LUIS application to identify low-confidence or misclassified utterances, which can then be reviewed and used to retrain the model for better accuracy.