Approaches to Data Labelling CT-AI Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A startup is developing a sentiment analysis model for social media posts. They have a limited budget and need to label a large dataset of public posts. They are considering using a crowdsourcing platform. What is the most important consideration to ensure the labels are reliable?

   • AUse a single annotator per item to keep costs low and avoid conflicting labels.
   • BUse only annotators who are native speakers of the language of the posts.
   • CPay annotators per item to incentivize speed, maximizing the number of labelled items.
   • DProvide clear labelling guidelines and use multiple annotators per item, then aggregate responses (e.g., majority vote).
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A company is using a crowdsourcing platform to label a dataset of images for a self-driving car project. The images contain pedestrians, vehicles, and traffic signs. What is a key challenge they are likely to face, and how can they mitigate it?

   • ASlow labelling speed; mitigate by paying annotators per image to incentivize speed.
   • BPrivacy concerns because the images contain people; mitigate by blurring faces before sharing.
   • CInconsistent labelling due to varying interpretations; mitigate by providing detailed guidelines and using multiple annotators per image.
   • DHigh cost of labelling; mitigate by using a pre-trained model to auto-label all images.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A company wants to label a large dataset of product reviews for sentiment (positive, negative, neutral). They have a small team of annotators and a tight deadline. They decide to use a pre-trained sentiment model to generate labels, then have the annotators review only the reviews where the model's confidence is below a threshold. What is the main benefit of this approach?

   • AIt ensures that the model's biases are removed from the labelling process.
   • BIt eliminates the need for human review entirely, saving time and money.
   • CIt reduces the amount of human labelling effort while maintaining quality by focusing human review on uncertain cases.
   • DIt guarantees that the labels are more accurate than manual labelling.
   Reveal Solution Discussion Select an answer first
5. 24foundation · medium

   Why is data labelling important for testing ML models?

   • AIt ensures the test data is free from any errors or noise
   • BIt makes the test data run faster on the hardware
   • CIt allows the model to be tested without any human involvement
   • DIt provides a ground truth against which the model's predictions can be compared to measure accuracy
   Reveal Solution Discussion Select an answer first
6. 25foundation · medium

   What is a primary benefit of using crowdsourcing for data labelling?

   • AIt is always the cheapest option for any labelling task
   • BIt guarantees high-quality labels because crowd workers are experts
   • CIt allows labelling tasks to be scaled quickly by distributing work to many people
   • DIt eliminates the need for labelling guidelines
   Reveal Solution Discussion Select an answer first