AI-102 Question 271
Select 3You are tasked with building a conversational AI solution for a customer service chatbot using Azure Language Understanding (LUIS). After training the model, you notice that it does not accurately predict intents for certain user inputs. To improve the model's performance, what steps should you take?
- A
Add more diverse examples for the intents in the training dataset.
- B
Deploy the model to production and rely on real-time feedback to improve performance.
- C
Use the Review endpoint to analyze misclassified predictions and retrain the model with corrected data.
- D
Reduce the number of intents in the model to simplify its understanding.
- E
Apply active learning by labeling suggested utterances provided by the system.
Show answer and explanation
Correct answers: A, C, E
Explanation
To improve the language understanding model's performance, you should focus on providing diverse and representative training data, leveraging tools like the Review endpoint to address misclassified predictions, and applying active learning to enrich the dataset with meaningful examples. These steps collectively enhance the model's ability to predict intents accurately and generalize to unseen inputs.
- A. Correct.
Adding more diverse examples for intents helps the model generalize better to various user inputs and improves its understanding, making this a valid step for improving performance.
- B. Incorrect.
Deploying the model to production without addressing the performance issues first is not recommended, as it may lead to poor user experience and inaccurate predictions.
- C. Correct.
Using the Review endpoint to analyze misclassified predictions and retraining the model with this information ensures that the model learns from its mistakes, making this an effective strategy.
- D. Incorrect.
Reducing the number of intents is not always the correct approach, as it may compromise the solution's functionality and does not directly address the issue of misclassification.
- E. Correct.
Applying active learning allows you to leverage the system's recommendations for unlabeled utterances, improving the model's training data and its overall performance.