Identify Features for Fake News Detection on Facebook
Overview
This question evaluates skills in feature engineering, model selection, and end-to-end machine-learning system design with attention to privacy-preserving user features, multimodal signals, and deployment constraints for flagging misinformation on a social platform.
Meta
Aug 4, 2025, 10:55 AM
Data Scientist
Onsite
Machine Learning
20
0
Design a Machine-Learning System to Flag Fake News on Facebook
Scenario
An increase in fake news has been detected on the platform. You are asked to design an automated system that flags likely misinformation for downranking, warning labels, or human review.
Task
Propose a feature set and modeling approach for a machine-learning system that flags fake news posts. Your answer should cover:
-
Features
- Text/content features
- User/account features (privacy-preserving, aggregated)
- Network/graph/cascade features
- Temporal/dynamic features
- Optional: link/source credibility and multimodal (image/video) signals
-
Modeling approach
- Baselines and production model(s)
- How to fuse heterogeneous signals
- Training data strategy (labels, noise, class imbalance)
-
Evaluation
- Offline metrics and validation protocol
- Thresholding, calibration, and cost trade-offs
- Fairness and robustness checks
-
Concept drift and ongoing monitoring
- Drift detection and retraining strategy
- Human-in-the-loop and active learning
State any assumptions you make and note practical deployment constraints (latency, language coverage, privacy).
Solution
Login required
Show