Answer-first summary for fast verification
Answer: Add synthetic training data where those phrases are used in non-toxic ways.
The correct answer is A: 'Add synthetic training data where those phrases are used in non-toxic ways.' This approach addresses the underlying issue of misclassification by enhancing the classifier's understanding of benign references to underrepresented religious groups. Adding synthetic training data is a cost-effective solution that can improve the model without needing significant resources, which is essential given the limited budget and overextended state of the team. Raising the threshold (Option D) would reduce false positives but could allow genuinely harmful comments to go unflagged, addressing the symptom rather than the cause.
Author: LeetQuiz Editorial Team
Ultimate access to all questions.
No comments yet.
Your organization manages an online message board that has seen a recent surge in toxic language and bullying. To combat this, you deployed an automated text classifier designed to flag and remove harmful comments. However, users are now reporting that some benign comments, particularly those referencing certain underrepresented religious groups, are being misclassified as abusive. A deeper analysis reveals that the classifier's false positive rate is disproportionately high for comments referencing these religious groups. Given your team's limited budget and current overextension, what should you do to address this issue?
A
Add synthetic training data where those phrases are used in non-toxic ways.
B
Remove the model and replace it with human moderation.
C
Replace your model with a different text classifier.
D
Raise the threshold for comments to be considered toxic or harmful.