← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

Senior
May 2026

Summary

Meta data scientist interview, one meaty ML evaluation question that covered a lot of ground fast. The kind of question where you think you know the answer and then realize halfway through you've forgotten to mention half the important stuff.

Questions Asked (1)

Q1

How would you design an offline evaluation framework for a video harm classifier? Walk through your choice of metrics, how you'd handle class imbalance, how ground truth gets collected, how you'd pick a classification threshold given business costs, and any fairness checks you'd run.

Product Analytics & MetricsTechnical Trade-offsSystem Design
Author's notes

This question sprawls in ways that are easy to underestimate.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the core components of an offline evaluation framework: metrics, class imbalance, ground truth, threshold selection, and fairness. Emphasize that the framework must align with business objectives and operational constraints, and discuss trade-offs explicitly.

Pro tip: Anchor your answer in the business context: harm classifiers often prioritize recall for severe harms, and thresholds should be set based on cost ratios. Mention that offline metrics are proxies and should be validated with online experiments.

1. Define Objectives and Metrics

Clarify the goal: minimize harmful content while controlling false positives. Choose metrics like precision, recall, F1, PR-AUC, and per-class recall, considering the cost of different error types.

2. Address Class Imbalance

Use techniques like stratified sampling, class weights, or synthetic data to handle imbalance. Evaluate with metrics robust to imbalance, such as PR-AUC and balanced accuracy.

3. Collect Ground Truth

Leverage human reviewers with clear guidelines and quality checks (e.g., multiple annotators, adjudication). Consider active learning to prioritize uncertain samples and ensure representative sampling.

4. Select Classification Threshold

Model the business costs of false positives and false negatives. Use cost-sensitive analysis to pick a threshold that minimizes expected cost, and validate with a holdout set.

5. Run Fairness Checks

Evaluate performance across demographic groups and content types. Check for disparate impact, equal opportunity, and other fairness metrics, and mitigate biases if needed.

Key Points to Mention

  • Precision-recall trade-off and cost-sensitive threshold selection
  • Handling class imbalance with resampling or class weights
  • Human-in-the-loop annotation with quality control
  • Fairness metrics like equal opportunity and disparate impact
  • Offline-online consistency and validation with A/B tests
  • Iterative refinement based on error analysis

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.