Manage LUIS Utterances
Question
You have created example utterances and intents for your LUIS application.
You train your application and while you review the dashboard, you identify certain issues.
Your top intent and the next intent scores are close enough that may flip with the next training.
To address the issue, you decide to delete a number of utterances amongst different intents.
This changes the quantity of example utterances significantly.
With this change, your review dashboard analysis again.
What kind of issue wouldyou expect on the dashboard with the new change?
Answers
Explanations
Click on the arrows to vote for the correct answer
A. B. C. D.Correct Answer: C.
Option A is incorrect because incorrect predictions occur when example utterance is not predicted for the labeled intent.
Instead, it is predicted for a different intent.
To remediate this issue, you would need to edit the utterances to be more specific and train the model again.
Option B is incorrect because unclear predictions occur when top intent and next intent scores are close enough to flip the results with the next model training.
To remediate this issue, you would need to combine the intents or edit the utterances and train the model again.
Option C is correct because data imbalance occurs when the quantity of example utterances varies significantly.
To remediate this issue, you would need to add more utterances to the intent and train the model again.
Reference:
To learn more about issues that can be fixed using dashboard analysis, use the link given below:
When training a Language Understanding Intelligent Service (LUIS) application, it is common to encounter issues in the review dashboard, particularly with the prediction scores of the different intents. In this scenario, the issue is that the top intent and the next intent scores are very close and may flip with the next training. This implies that LUIS is having difficulty distinguishing between the top two intents, and it may not accurately predict the correct intent for new utterances.
To address this issue, the decision is made to delete some utterances from different intents. This can have a significant impact on the quantity of the example utterances, and it is likely to lead to a data imbalance in the training data. The training data will no longer be representative of the actual data, which could lead to incorrect predictions or unclear predictions.
Therefore, the most likely issue that would be expected on the dashboard with the new change is C. Data imbalance. The data imbalance can negatively affect the performance of the LUIS application, as the model may not have enough data to accurately predict the intents.
It is important to note that deleting utterances should be done carefully, as it can have unintended consequences. It is best to keep a backup of the original data before making any changes, and to carefully monitor the performance of the application after any modifications.
