AI-102 Question 274
Select 3You are developing a customer support chatbot using Azure Language Understanding (LUIS). After deploying the language understanding model, you notice that the model is not correctly identifying intents for some key customer queries. What steps can you take to optimize the language understanding model for better performance?
- A
Add more representative utterances to the intents with low accuracy.
- B
Increase the number of entities in the model without verifying their relevance.
- C
Use the active learning feature to label suggestions for ambiguous queries.
- D
Analyze the model's performance using the LUIS endpoint logs and retrain the model.
- E
Reduce the number of intents to simplify the model, even if it means merging unrelated intents.
Show answer and explanation
Correct answers: A, C, D
Explanation
Optimizing a language understanding model involves adding representative utterances, leveraging active learning to address ambiguous queries, and analyzing endpoint logs to identify patterns for improvement. These steps ensure the model is refined based on real-world usage data, enhancing its accuracy and reliability. On the other hand, actions like adding irrelevant entities or merging unrelated intents can degrade the model's performance.
- A. Correct.
Adding representative utterances to intents with low accuracy helps the model better understand the variety of ways users might phrase their queries, improving accuracy.
- B. Incorrect.
Increasing the number of entities without verifying their relevance can clutter the model and lead to a decrease in accuracy rather than improving it.
- C. Correct.
Using the active learning feature allows you to identify and label ambiguous or misclassified queries, which is an effective way to refine the model.
- D. Correct.
Analyzing the LUIS endpoint logs helps identify patterns in misclassified queries and offers insights on retraining the model to improve its performance.
- E. Incorrect.
Reducing the number of intents by merging unrelated intents can confuse the model and lead to incorrect predictions, as intents should represent distinct tasks or goals.