Implement Custom Language Models AI-102 Practice Questions (Page 10)

Questions 46–50

1. 46application · medium

   A custom question answering knowledge base for a retail company has low confidence scores for many user questions. Users often phrase questions differently than the Q&A pairs in the knowledge base. What should the team do to improve the match rate?

   • AIncrease the confidence score threshold to filter out low-confidence matches.
   • BRemove Q&A pairs that are not frequently accessed.
   • CPublish the knowledge base and use user feedback to manually correct answers.
   • DAdd alternate phrasing to the Q&A pairs to cover common variations of the questions.
   Reveal Solution Discussion Select an answer first
2. 47application · medium

   A support team has a custom question answering project with a few QnA pairs. They want to improve the system's ability to answer questions about product returns. They have a document with detailed return policies. What should they do?

   • AImport the document as a source to automatically extract QnA pairs.
   • BManually add every possible question as a separate QnA pair.
   • CPublish the knowledge base to make the document available to users.
   • DTrain the knowledge base without adding new content.
   Reveal Solution Discussion Select an answer first
4. 48application · medium

   A company is building a custom question answering system for their IT helpdesk. They want the system to handle follow-up questions, such as asking for more details after a user reports a problem. What should they implement?

   • AMulti-turn conversation prompts that guide the user to provide additional information.
   • BAlternate phrasing for each QnA pair to cover different ways users might ask.
   • CPublish the knowledge base and rely on the user to rephrase their question.
   • DA single QnA pair with a long answer that covers all possible follow-ups.
   Reveal Solution Discussion Select an answer first
5. 49expert · medium

   A team is building a custom language model for a travel booking system. They have defined an intent 'BookFlight' and an entity 'DestinationCity'. The model is misclassifying 'I want to fly to Paris' as 'GetWeather' because the training data includes many weather-related utterances with city names. The team has limited time and cannot collect more data. What is the most effective way to reduce this misclassification?

   • AIncrease the number of entities to include more city names.
   • BAdd more utterances to the 'BookFlight' intent that include city names and clearly indicate booking intent, then retrain.
   • CLower the confidence threshold for the 'BookFlight' intent.
   • DRemove all weather-related utterances from the training data.
   Reveal Solution Discussion Select an answer first
6. 50foundation · easy

   In Azure AI Language, what is a common way to back up a custom language model?

   • ADelete the model and recreate it
   • BExport the project to a JSON file
   • CTrain the model again
   • DChange the model's name
   Reveal Solution Discussion Select an answer first