AI-102 Question 270
Select 4You are developing a customer support chatbot using Azure Language Understanding (LUIS). The chatbot needs to identify user intents such as 'CreateTicket', 'CheckTicketStatus', and 'CancelTicket'. After training the model, you notice that the chatbot often misclassifies 'CancelTicket' intents as 'CheckTicketStatus'. Which of the following actions should you take to improve the model's performance?
- A
Add more example utterances for the 'CancelTicket' intent.
- B
Ensure there are no overlapping utterances between 'CancelTicket' and 'CheckTicketStatus' intents.
- C
Increase the number of intents in the model to make it more specific.
- D
Use the Test feature in LUIS to analyze and refine the model's predictions.
- E
Enable Active Learning in LUIS to identify and correct misclassified utterances.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To improve a language understanding model's performance, it's essential to provide sufficient example utterances, avoid overlapping utterances between intents, and iteratively test and refine the model. In this scenario, adding more utterances, ensuring no overlaps, testing the model, and enabling Active Learning are all effective measures to address the misclassification issue. Increasing the number of intents is not a valid solution as it does not address the root cause of the problem.
- A. Correct.
Adding more example utterances for the 'CancelTicket' intent will help the model better understand and differentiate this intent.
- B. Correct.
Overlapping utterances between intents can confuse the model. Ensuring no overlap will improve intent classification accuracy.
- C. Incorrect.
Increasing the number of intents might dilute the model's focus and is not directly related to resolving the misclassification issue.
- D. Correct.
Using the Test feature will allow you to evaluate and refine the model's performance by checking how it predicts various utterances.
- E. Correct.
Enabling Active Learning helps identify frequently misclassified utterances, allowing you to provide corrections and improve the model's overall accuracy.