AI-102 Question 278
Select 3You are tasked with optimizing a language understanding model built using Azure Language Understanding (LUIS). The model is performing well on common user inputs but struggles with less frequent intents and entities. Which of the following actions should you take to improve the model's performance?
- A
Add more labeled examples for the underperforming intents and entities.
- B
Use the Batch Testing feature in LUIS to analyze model predictions on a test dataset.
- C
Reduce the number of intents in the model to improve prediction accuracy for all intents.
- D
Enable Active Learning to identify and label ambiguous user utterances.
- E
Increase the threshold for intent confidence scores to filter out low-confidence predictions.
Show answer and explanation
Correct answers: A, B, D
Explanation
To optimize a language understanding model in Azure LUIS, you need to focus on improving its training data and evaluation process. Adding more labeled examples for underperforming intents and entities ensures better model training. Batch Testing provides a way to identify specific weaknesses in the model's predictions. Active Learning helps in iteratively improving the model by highlighting ambiguous cases. These actions collectively enhance the model's performance without compromising its functionality.
- A. Correct.
Adding more labeled examples helps the model better understand less frequent intents and entities, improving its ability to generalize.
- B. Correct.
Batch Testing allows you to evaluate the model's performance on a test dataset, helping you identify specific areas where it needs improvement.
- C. Incorrect.
While reducing the number of intents may simplify the model, it is not always necessary or desirable, as it may compromise the application's functionality.
- D. Correct.
Active Learning identifies ambiguous utterances that the model struggles with, allowing you to label them and improve the model iteratively.
- E. Incorrect.
Increasing the threshold for intent confidence scores does not directly improve the model's optimization but merely filters predictions, which could lead to missed intents.