Design: Feed Ranking quiz

Question 1

Which objective function best captures long-term user satisfaction in a feed ranking model?

Accepted Answer

A multi-task objective combining short-term engagement and long-term retention signals

Answer

Maximizing raw click-through rate on all feed items

Answer

Minimizing time spent per session to improve user efficiency

Answer

Maximizing the number of posts shown per session regardless of engagement

Question 2

What is the primary challenge with using dwell time as a training signal for feed ranking?

Accepted Answer

Dwell time is a delayed label: it is only finalized after the user navigates away

Answer

Dwell time is too expensive to measure at billion-user scale

Answer

Dwell time is always a perfect proxy for user satisfaction

Answer

Dwell time data cannot be stored in real-time feature stores

Question 3

What is 'bandit feedback' in the context of feed ranking?

Accepted Answer

Partial reward observability—we only see engagement for posts the system chose to show

Answer

Feedback from users who click on ads injected into the feed

Answer

Feedback collected from A/B tests with randomly assigned users to treatment groups

Answer

Signals from users who have been banned or flagged for policy violations

Question 4

What distinguishes feed ranking from search ranking in system design?

Accepted Answer

Feed ranking is proactive with no explicit query; search ranking responds to explicit user intent

Answer

Feed ranking uses a transformer model; search ranking uses BM25

Answer

Feed ranking requires more compute than search ranking in all cases

Answer

Feed ranking only applies to social networks while search applies to all platforms

Question 5

What is the 'rich-get-richer' problem in feed ranking?

Accepted Answer

High-engagement posts receive more impressions, generate more engagement, and further dominate ranking

Answer

Popular posts always deserve to be ranked higher because users genuinely prefer them

Answer

The system intentionally promotes paid content over organic content

Answer

Rich-get-richer refers to advertisers outcompeting organic content in auctions

Question 6

Which feature is most important for real-time feed personalization within an active session?

Accepted Answer

The user's recent engagement signal within the current session

Answer

The user's long-term demographic profile stored in the user database

Answer

The global trending score of each post across all users

Answer

The creator's average historical engagement rate over the past 30 days

Question 7

How does exploration-exploitation work differently in feed ranking versus search ranking?

Accepted Answer

Feed exploration has session-level effects and must be position-managed and quality-gated

Answer

Feed ranking explores per-query just like search ranking does

Answer

Exploration is only relevant in search ranking, not feed ranking

Answer

Feed exploration always uses random sampling while search uses UCB

Question 8

What is 'position bias' in the context of a vertically scrolling feed?

Accepted Answer

The observation that posts near the top of the feed receive more engagement due to higher visibility

Answer

The tendency for users to follow accounts that appear in sponsored positions

Answer

The ranking model's bias toward content from accounts with many followers

Answer

The system's preference for promoted content over organic content

Question 9

What does 'soft labels' mean in feed ranking model training?

Accepted Answer

Probabilistic or continuous relevance scores used as training targets instead of binary labels

Answer

Hard decision boundaries assigned by human annotation teams

Answer

Labels generated by automatically running A/B tests on each post

Answer

Labels derived from advertisers' bidding signals for organic feed content

Question 10

Which metric best measures feed ranking quality for long-term user retention?

Accepted Answer

N-day return rate measuring what fraction of users come back after a session

Answer

Raw click-through rate averaged across all sessions and users

Answer

Total number of posts shown per session without regard to engagement quality

Answer

Average session duration without accounting for passive versus active engagement

Question 11

What is the 'engagement trap' in feed ranking?

Accepted Answer

Optimizing for short-term engagement leads to harmful content that drives long-term churn

Answer

The system becoming stuck recommending the same content repeatedly to a user

Answer

Users becoming trapped in paid subscription models through feed manipulation

Answer

The technical trap of infinite feed scroll increasing infrastructure costs

Question 12

How do you handle position bias correction in feed ranking training data?

Accepted Answer

Apply IPS or examination hypothesis correction using position propensities estimated from randomization experiments

Answer

Only use clicks from the top position since they are the least biased

Answer

Remove all positional information from the training set before model training

Answer

Collect more data to overcome position bias through scale alone

Question 13

What should trigger a feed ranking model retrain?

Accepted Answer

Any combination of metric degradation, feature drift, content shifts, or scheduled cadence

Answer

Only when the engineering team identifies a new model architecture to deploy

Answer

Only after a full human audit of the current model's predictions across all content types

Answer

Retraining is not needed if the model uses any form of online learning or bandit updates

Test yourself