Question 14
Domain 4: Implement Natural Language Processing SolutionsA company's CLU model correctly identifies intents during testing but performs poorly in production where users phrase things differently than the training utterances. What is the most effective improvement strategy?
Correct answer: B
Explanation
The model is overfitting to the training phrasing, so it needs broader intent coverage. Adding diverse utterances, especially "natural variations from production logs," improves generalization because the CLU model learns the different ways users express the same intent.
Why each option is right or wrong
A. Increase the confidence threshold to reduce false positives
B. Add diverse utterances to each intent, including natural variations from production logs
Azure AI Language Conversational Language Understanding (CLU) improves from supervised training data, so when test accuracy is high but live traffic fails on paraphrased user wording, the issue is insufficient utterance coverage rather than model selection. The remedy is to expand each intent with many labeled examples that reflect real production phrasing; in practice, Microsoft recommends using varied, representative utterances from actual logs so the model can generalize beyond the exact training wording.
C. Enable entity linking in the model
D. Deploy the model to a new region