Design a system to evaluate the credibility of news articles using machine learning.
Instruction: Detail the features you would consider and the architecture of the system.
Updated
Example Answer
I would treat this as a system for checking claims and helping reviewers, rather than a model that decides whether an entire article is true. I would extract checkable claims, retrieve relevant evidence, and keep the evidence and its date with each assessment. If the evidence is missing or contradictory, the system should show that uncertainty.
I would start with a narrowly defined subject area and human-reviewed examples. The pipeline would separate retrieval quality from the claim assessment so we could tell whether it failed to find evidence or misread what it found. I would evaluate false accusations, missed misleading claims, and unsupported decisions, with a route for review and correction. Writing style, sentiment, and publisher reputation may provide context, but they cannot establish that a particular claim is false.
Make it your own
Choose the subject area, evidence sources, and review process. Explain how opinion, satire, outdated information, and a genuinely checkable claim would be handled. Do not imply that releasing a dataset is automatically appropriate when it contains restricted or sensitive material.
Why this works
It separates evidence-based verification from shortcuts and builds in uncertainty and review.
Follow-up to practice
What should the system show when reliable sources disagree?