← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
Jun 2026Remote

Summary

Meta DS interview that went deep into ML evaluation for a fake account detection problem. The whole thing was basically one long case question with a bunch of sub-parts, and I was not fully prepared for how much math they expected me to do on the spot.

Questions Asked (5)

Q1

You suspect fake users are inflating comment counts. You need to build a classifier to flag them for review. Given severe class imbalance and a hard daily review capacity, which offline evaluation metrics would you choose and why? When would you prefer PR-AUC over ROC-AUC?

Product Analytics & MetricsTechnical Trade-offs
Author's notes

This is where I spent the most time and probably where I lost the most points.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the problem: severe class imbalance and a hard daily review capacity mean evaluation must focus on precision at the top of the ranked list. Explain why PR-AUC is more informative than ROC-AUC when the negative class dominates, and tie metrics to the operational constraint by recommending precision@k or recall@k for a fixed review budget. Conclude with when ROC-AUC might still be useful, such as when ranking across the full score range matters or when the positive class is not extremely rare.

Pro tip: Emphasize that the choice of metric should align with the business cost of false positives (wasted reviewer time) and false negatives (missed fake accounts), and propose a cost-sensitive evaluation that reflects the daily review capacity.

1. Clarify the operational constraint

Restate that the team can only review a fixed number of comments per day, so the classifier must prioritize the most likely fake users. This makes top-k precision the primary business metric.

2. Explain why PR-AUC is preferred under imbalance

PR-AUC focuses on the positive (fake) class and is not inflated by the large number of true negatives. ROC-AUC can look optimistic because the false positive rate remains low even with many false positives when negatives dominate.

3. Recommend specific offline metrics

Suggest precision@k (where k equals daily review capacity), recall@k, and PR-AUC. Also consider F-beta with beta < 1 if false positives are costlier, or cost-sensitive metrics.

4. Discuss when ROC-AUC might still be useful

ROC-AUC is appropriate when the positive and negative classes are more balanced, when you care about ranking across the entire score range, or when the decision threshold is not fixed. But under severe imbalance, it can be misleading.

5. Tie back to business impact

Conclude that the chosen metrics should directly reflect the trade-off between catching fake users and wasting reviewer time, and suggest monitoring these metrics in production as the class distribution shifts.

Key Points to Mention

  • Class imbalance: PR-AUC is more sensitive to changes in the positive class and avoids the misleadingly high ROC-AUC.
  • Precision@k and recall@k: directly map to the daily review capacity and business goal.
  • Cost-sensitive evaluation: false positives waste reviewer time, false negatives let fake users inflate counts.
  • ROC-AUC vs PR-AUC: ROC-AUC can be optimistic when negatives dominate; PR-AUC focuses on the minority class.
  • Threshold selection: with a hard capacity, choose a threshold that yields exactly k predictions per day.
  • Baseline comparison: compare PR-AUC to the no-skill baseline (positive rate) to assess true lift.

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

Q2

Given a population of 10 million daily active users with a 1% true fake rate, compute the expected true positives and false positives per day for two models: Model A with precision 0.60 and recall 0.20, and Model B with precision 0.20 and recall 0.80. Does either model fit within a 50,000 account/day review capacity?

Product Analytics & MetricsTechnical Trade-offs
Author's notes

The math part.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, compute the total number of fake accounts (1% of 10M = 100,000) and real accounts (9.9M). Then, for each model, use precision and recall to derive the confusion matrix: true positives = recall * actual fakes; false positives = (true positives / precision) - true positives. Finally, compare the total flagged accounts (TP + FP) to the 50,000 review capacity.

Pro tip: Emphasize that precision and recall alone don't determine capacity fit; you must compute the absolute numbers. Also, note that Model B's high recall but low precision may overwhelm the review team, while Model A's high precision but low recall misses many fakes—highlight the trade-off and suggest that the optimal model depends on business priorities (e.g., minimizing missed fakes vs. review cost).

1. Calculate base rates

Compute the total number of fake and real accounts: 1% of 10M = 100,000 fakes; 9.9M real accounts.

2. Compute true positives for each model

Use recall to find true positives: TP = recall * actual fakes. For Model A: 0.20 * 100,000 = 20,000. For Model B: 0.80 * 100,000 = 80,000.

3. Compute false positives for each model

Use precision to find total predicted positives: total predicted = TP / precision. Then FP = total predicted - TP. For Model A: total predicted = 20,000 / 0.60 ≈ 33,333; FP ≈ 13,333. For Model B: total predicted = 80,000 / 0.20 = 400,000; FP = 320,000.

4. Compare to review capacity

Total flagged accounts = TP + FP. Model A: 20,000 + 13,333 = 33,333 (within 50,000). Model B: 80,000 + 320,000 = 400,000 (exceeds 50,000).

5. Interpret trade-offs

Discuss that Model A fits capacity but misses 80% of fakes; Model B catches more fakes but requires 8x the review capacity. Suggest potential adjustments like threshold tuning or combining models.

Key Points to Mention

  • Precision = TP / (TP + FP), so FP = (TP / precision) - TP.
  • Recall = TP / (TP + FN), so TP = recall * actual positives.
  • Total flagged accounts = TP + FP, which must be ≤ review capacity.
  • Model A: TP=20k, FP≈13.3k, total≈33.3k (fits). Model B: TP=80k, FP=320k, total=400k (does not fit).
  • Trade-off: Model A has high precision but low recall (misses many fakes); Model B has high recall but low precision (many false alarms).
  • Business implication: If missing fakes is costly, Model B might be preferred despite capacity issues; if review resources are limited, Model A is more feasible.

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

Q3

Given that a false positive costs $2 (review cost) and a false negative costs $100 (missed abuse), which Fβ score would you choose and why? Calculate the expected daily cost under both models at their current thresholds.

Product Analytics & MetricsTechnical Trade-offs
Author's notes

β squared equals the ratio of how much more you care about recall vs precision, so FN/FP cost ratio is 50, meaning β = sqrt(50) which is about 7.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, translate the asymmetric costs into a cost ratio (100:2 = 50:1) and connect it to the Fβ score by recalling that β² equals the ratio of false negative cost to false positive cost. Then, compute the expected daily cost for each model using the confusion matrix counts and compare them to justify your choice of β and threshold.

Pro tip: Emphasize that the optimal threshold depends on the cost ratio, not just the Fβ score; you can derive the threshold that minimizes expected cost by setting the odds ratio equal to the cost ratio. This shows you understand the decision-theoretic foundation.

1. Translate costs to β

Compute β from the cost ratio: β = sqrt(C_FN / C_FP) = sqrt(100/2) = sqrt(50) ≈ 7.07. This indicates that recall is about 7 times more important than precision.

2. Choose Fβ score

Select Fβ with β ≈ 7 (or F7) because it weights recall much higher, aligning with the higher cost of false negatives. Explain that F1 would be inappropriate due to asymmetric costs.

3. Calculate expected daily cost

For each model, use the confusion matrix at its current threshold: Expected Cost = (FP * $2) + (FN * $100). Sum these to get the total daily cost per model.

4. Compare and decide

Compare the expected costs of both models. The model with lower expected cost is preferable, even if its Fβ score is not the highest, because it directly minimizes the business cost.

5. Discuss threshold optimization

Mention that the current thresholds may not be optimal; you could adjust them to minimize expected cost further, using the cost ratio to set the decision threshold.

Key Points to Mention

  • β² = C_FN / C_FP, so β = sqrt(100/2) ≈ 7.07, meaning recall is ~7x more important than precision.
  • Fβ score with β > 1 emphasizes recall, which is appropriate when false negatives are costlier.
  • Expected cost formula: Total Cost = FP * C_FP + FN * C_FN.
  • The optimal decision threshold is where the likelihood ratio equals the cost ratio (C_FP / C_FN).
  • Compare models based on expected cost, not just Fβ, because Fβ is a proxy for cost-sensitive performance.
  • Consider that the current thresholds might not be cost-optimal; you can tune them to minimize expected cost.

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

Q4

How would you choose a classification threshold using a precision-recall curve, subject to a constraint of either precision at or above 0.7 or fewer than 20,000 false positives per day? How does probability calibration fit into this process?

Technical Trade-offsA/B Testing & Experimentation
Author's notes

Talked through sweeping thresholds along the PR curve and finding the lowest threshold that satisfies the precision constraint, since lower thresholds give more recall.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business context and the trade-off between precision and false positives, then walk through how to use the precision-recall curve to select a threshold that satisfies the constraint. Emphasize that calibration ensures predicted probabilities are meaningful, so threshold selection based on probabilities is valid.

Pro tip: Always validate the chosen threshold on a holdout set and monitor its performance over time, as data drift can shift the precision-recall curve. Also, consider the cost of false positives versus false negatives to ensure the threshold aligns with business objectives.

1. Understand the constraints and business context

Clarify whether the constraint is precision ≥ 0.7 or false positives < 20,000 per day, and discuss the implications of each. Determine the cost of false positives and false negatives to guide the trade-off.

2. Generate and evaluate the precision-recall curve

Use a validation set to compute precision and recall at various thresholds. Plot the precision-recall curve to visualize the trade-off and identify the feasible region that satisfies the constraint.

3. Select a threshold that meets the constraint

If the constraint is precision ≥ 0.7, find the threshold where precision is at least 0.7 and recall is maximized. If the constraint is false positives < 20,000 per day, estimate the daily volume and choose a threshold that keeps false positives below that limit while maximizing recall.

4. Incorporate probability calibration

Ensure the model's predicted probabilities are well-calibrated (e.g., using Platt scaling or isotonic regression) so that threshold selection based on probabilities is reliable. Calibration helps in interpreting the threshold and comparing across models.

5. Validate and monitor the threshold

Evaluate the chosen threshold on a holdout test set to confirm it meets the constraint. Set up monitoring to detect drift and re-evaluate the threshold periodically.

Key Points to Mention

  • Precision-recall curve is more informative than ROC when dealing with imbalanced classes.
  • Threshold selection depends on the specific constraint: precision floor vs. false positive cap.
  • Probability calibration ensures that predicted probabilities reflect true likelihoods, making threshold selection meaningful.
  • False positives per day requires estimating daily prediction volume and converting the constraint to a rate.
  • Trade-off between precision and recall: maximizing recall subject to precision constraint or false positive limit.
  • Use of validation set to avoid overfitting the threshold and holdout set for final evaluation.

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

Q5

Describe a validation approach that avoids data leakage for this kind of temporal abuse detection problem. What offline-to-online guardrails would you use, and which online metrics would you track post-launch?

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

Time-based splits, not random k-fold, because random splits leak future behavior into training.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by emphasizing that temporal abuse detection requires time-aware validation to prevent leakage, such as using a rolling-origin or expanding-window time series split. Then outline offline-to-online guardrails like shadow deployment and canary testing, and finish with online metrics that monitor both model performance and business impact.

Pro tip: Highlight that leakage often comes from feature engineering (e.g., using future data in aggregates) and that you'd implement a feature availability check to ensure each feature is computable at prediction time. Also, mention that you'd track delayed feedback metrics to capture abuse that manifests later.

1. Time-aware validation

Use a temporal split (e.g., train on past, validate on future) with a gap to avoid leakage from label delay. Consider expanding-window cross-validation to simulate real deployment.

2. Feature leakage audit

Audit features for temporal leakage: ensure all features are computed only from data available before the prediction timestamp. Use point-in-time correctness checks.

3. Offline-to-online guardrails

Before full launch, run shadow mode to compare model predictions with current system without affecting users. Then use canary testing with a small traffic percentage and monitor for anomalies.

4. Online metrics tracking

Track model performance metrics (precision, recall, AUC) and business metrics (abuse rate, user reports, revenue impact). Also monitor data drift and feature distribution shifts.

5. Iterative monitoring and rollback

Set up automated alerts for metric degradation and a rollback plan. Continuously retrain with new data and re-validate to adapt to evolving abuse patterns.

Key Points to Mention

  • Temporal validation techniques: rolling-origin, expanding-window, and gap between train and validation to handle label delay.
  • Feature leakage prevention: point-in-time correctness, avoiding future aggregates, and using only past data for feature computation.
  • Offline-to-online guardrails: shadow deployment, canary testing, and A/B testing with holdout groups.
  • Online metrics: model performance (precision, recall, AUC), business metrics (abuse rate, user reports, revenue), and system metrics (latency, throughput).
  • Monitoring for data drift and concept drift, with automated alerts and retraining pipelines.
  • Handling delayed feedback: using proxy metrics or delayed labels to evaluate model performance in the short term.

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