← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Meta DS interview focused entirely on ML evaluation for a recommendation system, specifically logistic regression scoring for restaurant recs. Five questions, all technical, no behavioral fluff. Pretty dense for a single session.

Questions Asked (5)

Q1

How would you design a defensible offline evaluation protocol to compare two recommendation models without access to A/B tests? Walk through the metrics you'd use, how you'd prevent data leakage, how you'd pick K for ranking metrics, and name a pitfall of inverse propensity scoring when propensities get small.

A/B Testing & ExperimentationProduct Analytics & MetricsTechnical Trade-offs
Author's notes

This was the hardest one.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around a clear offline evaluation pipeline: start with data splitting and leakage prevention, then define metrics and choose K, and finally address advanced topics like IPS and its pitfalls. Emphasize that offline evaluation is a proxy for online performance, so you must carefully design to avoid biases and ensure defensibility.

Pro tip: Mention that you would validate your offline protocol by checking correlation with historical A/B tests or by running a small online experiment to calibrate, showing you understand the limitations and the need for continuous validation.

1. Data Splitting and Leakage Prevention

Use time-based splitting to simulate real-world deployment, ensuring no future data leaks into training. Consider user-level splitting if interactions are user-specific, and remove any features that would not be available at inference time.

2. Metric Selection and K Choice

Choose metrics that align with business goals, such as precision@K, recall@K, NDCG, or MAP. Pick K based on the number of items users typically interact with (e.g., top-10 for a homepage) and consider multiple K values to assess robustness.

3. Handling Bias and Counterfactuals

Address selection bias by using techniques like inverse propensity scoring (IPS) or doubly robust estimation. Be aware of IPS pitfalls when propensities are small, such as high variance and instability, and consider clipping or using SNIPS.

4. Validation and Sensitivity Analysis

Validate the offline protocol by checking correlation with historical A/B tests if available. Perform sensitivity analysis on K, metrics, and model hyperparameters to ensure conclusions are robust.

Key Points to Mention

  • Time-based splitting to prevent temporal leakage
  • Use of ranking metrics like NDCG, MAP, and precision@K
  • Choosing K based on user engagement patterns and business context
  • Inverse propensity scoring (IPS) and its variance issue with small propensities
  • Doubly robust estimation as a more stable alternative to IPS
  • Validation of offline metrics against online A/B test results

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

Q2

Given a holdout set with specific TP, FP, TN, FN counts at a 0.7 threshold, compute precision, recall, specificity, F1, and accuracy. Then explain why accuracy is misleading in this context and which metric best fits the goal of only showing relevant items.

Product Analytics & MetricsTechnical Trade-offs
Author's notes

Straightforward math, no issues there.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly writing out the confusion matrix and the formulas for each metric, then compute them step by step. After obtaining the numbers, interpret them in the context of the business goal—likely minimizing false positives to ensure only relevant items are shown—and explain why accuracy is misleading due to class imbalance. Conclude by recommending precision as the most appropriate metric, possibly with a trade-off discussion.

Pro tip: Always relate the metrics back to the product goal: for a system that shows only relevant items, false positives are costly (user annoyance), so precision is key. Mention that you might also consider precision@k or a weighted metric if the cost of FP is much higher than FN.

1. Define the confusion matrix and metrics

Write down the given TP, FP, TN, FN counts and state the formulas for precision, recall, specificity, F1, and accuracy.

2. Compute each metric

Plug in the numbers and calculate precision, recall, specificity, F1, and accuracy. Show your work to demonstrate rigor.

3. Explain why accuracy is misleading

Discuss class imbalance: if negatives dominate, a model predicting all negatives can have high accuracy but fail to identify positives. Accuracy doesn't distinguish between types of errors.

4. Identify the best metric for the goal

Argue that precision is most relevant because the goal is to show only relevant items (minimize false positives). Optionally, mention F1 if both precision and recall are important, but precision aligns with the stated objective.

5. Discuss trade-offs and additional considerations

Acknowledge that optimizing precision may reduce recall, and suggest monitoring both. Mention threshold tuning and business impact of FP vs FN.

Key Points to Mention

  • Formulas: Precision = TP/(TP+FP), Recall = TP/(TP+FN), Specificity = TN/(TN+FP), F1 = 2*(Precision*Recall)/(Precision+Recall), Accuracy = (TP+TN)/(TP+TN+FP+FN)
  • Class imbalance and its effect on accuracy
  • Precision as the metric that directly addresses false positives
  • The trade-off between precision and recall
  • Threshold selection and its impact on metrics
  • Business context: cost of false positives vs false negatives

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

Q3

How would you check and fix probability calibration in a logistic regression model? Why does calibration matter specifically when you're using a score threshold to make business decisions?

Product Analytics & MetricsTechnical Trade-offs
Author's notes

Reliability diagrams came to mind immediately.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining calibration and explaining why it matters for threshold-based decisions, then walk through diagnostic methods (reliability diagrams, calibration curves, Brier score) and fixes (Platt scaling, isotonic regression, or retraining with proper loss). Emphasize that calibration ensures predicted probabilities reflect true likelihoods, which is critical when choosing a threshold to balance business costs and benefits.

Pro tip: Always validate calibration on a holdout set and consider the business impact of miscalibration—e.g., if you threshold at 0.5 but the model is overconfident, you may reject good customers or approve bad ones, directly affecting revenue and risk.

1. Define calibration and its importance

Explain that a calibrated model's predicted probabilities match observed frequencies (e.g., among predictions of 0.7, about 70% should be positive). Highlight that calibration is separate from discrimination (AUC) and crucial for threshold-based decisions.

2. Diagnose calibration

Use reliability diagrams (calibration curves) to visualize predicted vs. observed probabilities, and compute metrics like Brier score or log loss. Also check for overconfidence/underconfidence by binning predictions and comparing to actual outcomes.

3. Apply calibration fixes

For post-hoc calibration, use Platt scaling (logistic regression on scores) or isotonic regression (non-parametric, more flexible but needs more data). Alternatively, retrain with a proper scoring rule (e.g., log loss) and ensure no data leakage.

4. Validate and monitor

Evaluate calibration on a holdout set using reliability diagrams and metrics. Implement ongoing monitoring because calibration can drift over time as data distributions change.

5. Connect to business decisions

Explain that when using a threshold (e.g., approve if p > 0.5), miscalibration directly impacts the expected cost/benefit. For example, if the model is overconfident, you might approve too many risky cases; if underconfident, you might reject profitable ones.

Key Points to Mention

  • Reliability diagram (calibration curve) as the primary diagnostic tool
  • Platt scaling and isotonic regression as common post-hoc calibration methods
  • Brier score and log loss as proper scoring rules that assess calibration
  • The distinction between discrimination (AUC) and calibration
  • The impact of calibration on threshold-based decisions: false positives/negatives and business costs
  • Need for a holdout set and monitoring for calibration drift

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

Q4

Why would you choose logistic regression over a more complex model for this recommendation scoring task? Describe two failure modes of logistic regression and how you'd address them.

Technical Trade-offsSystem Design
Author's notes

Went with interpretability, fast iteration, and the fact that calibration is more natural out of the box.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing logistic regression as a strong baseline that offers interpretability, speed, and robustness for recommendation scoring, especially when data is limited or latency matters. Then, acknowledge its limitations by describing two failure modes and concrete mitigation strategies, showing you understand trade-offs and can make pragmatic choices.

Pro tip: Emphasize that at Meta's scale, a simple model that can be trained quickly and deployed with low latency often beats a complex model that's marginally better offline but harder to maintain. Mention that logistic regression can be a stepping stone to more complex models, and you'd only upgrade if the lift justifies the cost.

1. Justify logistic regression

Explain why logistic regression is suitable: interpretability, fast training/inference, low latency, works well with sparse features, and provides calibrated probabilities. Highlight that it's a strong baseline for recommendation scoring.

2. Identify failure mode 1: non-linear relationships

Describe how logistic regression assumes linearity in the log-odds, so it may underperform when feature interactions or non-linear patterns are important. Propose solutions like feature engineering (e.g., crosses, bucketization) or using a more complex model if needed.

3. Identify failure mode 2: class imbalance or rare events

Explain that logistic regression can be biased toward the majority class when positive events (e.g., clicks) are rare. Suggest techniques like resampling, class weighting, or using precision-recall AUC for evaluation.

4. Address other potential issues

Briefly mention other failure modes like multicollinearity or outliers, and how to handle them (regularization, robust scaling). Show awareness of model diagnostics.

5. Conclude with trade-off mindset

Summarize that the choice depends on the problem constraints: if interpretability and speed are critical, logistic regression is preferred; if performance gains from complex models are significant, consider upgrading, but always start simple.

Key Points to Mention

  • Interpretability and explainability for debugging and stakeholder trust
  • Low latency and scalability for real-time recommendation serving
  • Feature engineering (e.g., interactions, binning) to capture non-linearity
  • Handling class imbalance via weighting, resampling, or appropriate metrics
  • Regularization (L1/L2) to prevent overfitting and handle multicollinearity
  • Baseline-first approach: start simple, then iterate to complex models if needed

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

Q5

If your model uses features derived from friend activity, how do you design your train/test split to avoid interference leakage? What are the tradeoffs between time-based, user-disjoint, and graph-clustered splits?

A/B Testing & ExperimentationTechnical Trade-offsSystem Design
Author's notes

Graph-clustered splits are the right answer for network interference but I had to think out loud for a bit before getting there.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the leakage problem: friend-derived features create dependencies between training and test examples, so random splits overestimate performance. Then compare time-based, user-disjoint, and graph-clustered splits on how well they mimic production, their bias-variance tradeoffs, and their impact on evaluation metrics.

Pro tip: Emphasize that the right split depends on the deployment scenario: if the model will be used on new users, user-disjoint is essential; if it will be used on future interactions of existing users, time-based is more appropriate. Also mention that graph-clustered splits can better preserve community structure but may reduce training data diversity.

1. Define the leakage problem

Explain that friend-derived features (e.g., friend embeddings, activity counts) create dependencies between examples, so random splits leak information and inflate test performance.

2. Describe split strategies

Outline time-based (split by timestamp), user-disjoint (split by user ID), and graph-clustered (split by community detection) splits, and how each prevents leakage differently.

3. Analyze tradeoffs

Compare the splits on realism, bias, variance, and computational cost. For example, time-based mimics temporal drift but may not generalize to new users; user-disjoint tests generalization to unseen users but may ignore temporal patterns; graph-clustered preserves community structure but can be complex to implement.

4. Recommend based on use case

Suggest which split to use depending on the production scenario (e.g., new user vs. existing user, real-time vs. batch) and mention that a combination (e.g., time-based within user-disjoint) might be needed.

5. Validate with metrics

Propose evaluating the model on multiple splits to understand performance sensitivity and to detect leakage, and using online A/B tests as the ultimate validation.

Key Points to Mention

  • Leakage from friend features: training and test examples share friends, causing information bleed.
  • Time-based split: respects temporal order, mimics production for existing users, but may not test new-user generalization.
  • User-disjoint split: ensures no user overlap, tests generalization to new users, but may ignore temporal dynamics and reduce training data.
  • Graph-clustered split: partitions graph into communities, reduces cross-community leakage, but can be computationally expensive and may not reflect production if communities are not isolated.
  • Tradeoffs: bias-variance, realism vs. control, computational complexity, and impact on evaluation metrics.
  • Best practice: choose split based on deployment scenario, validate with multiple splits, and use online experiments for final validation.

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