AI-102 Question 267
Select 3You are building a conversational AI solution using Azure Language Service and need to create entities for a custom Named Entity Recognition (NER) model. The goal is to extract specific information such as product names, order IDs, and customer names from customer support chat transcripts. Which steps should you follow to create and train entities for this use case?
- A
Define the entity categories such as 'ProductName', 'OrderID', and 'CustomerName' in your project.
- B
Manually annotate sample data by tagging phrases corresponding to the defined entities in the chat transcripts.
- C
Deploy the model directly without training, as Azure automatically recognizes entity patterns.
- D
Train the model using the annotated data to improve entity recognition accuracy.
- E
Export the trained model to a third-party platform for additional customization.
Show answer and explanation
Correct answers: A, B, D
Explanation
To create custom entities for a Named Entity Recognition (NER) model in Azure Language Service, you must define the specific categories of entities you want to extract, annotate sample data to train the model, and then train the model using the annotated data. This ensures that the model can accurately recognize the desired entities in customer support chat transcripts. Deploying without training or exporting the model to a third-party platform is not relevant to successfully implementing this Azure AI solution.
- A. Correct.
Defining entity categories is a crucial first step in setting up a custom NER model. It allows you to specify what information you want the model to extract.
- B. Correct.
Manually annotating sample data is necessary to teach the model how to recognize entities in context, ensuring it learns from real-world examples.
- C. Incorrect.
Deploying the model without training would not work in this case, as the model requires annotated data to learn how to extract custom entities.
- D. Correct.
Training the model using the annotated data is essential to improve its accuracy and ensure it can correctly identify entities based on the provided examples.
- E. Incorrect.
Exporting the trained model to a third-party platform is not required to implement a solution in Azure and is irrelevant to this specific use case.