hn.today

Ask HN: Should Uber Ratings Decay over Time?

news.ycombinator.com10 points5 comments
Screenshot of Ask HN: Should Uber Ratings Decay over Time?

The discussion questions whether Uber’s practice of calculating passenger ratings from the last 500 rated trips, without regard to when they happened, gives a fair picture of a rider’s current behavior. The writer has 455 ratings, 426 of them five stars, and an overall score of 4.86, but argues that the methodology matters more than the score: a poor rating from five years ago can count as much as one from yesterday. For people who ride infrequently and have fewer than 500 ratings, old negative feedback may continue affecting their reputation for years. The underlying tension is whether a reputation score should preserve a person’s complete history or represent who they are now.

One proposed alternative is exponential time decay, with each rating’s weight halving after a set period - for example, a 12-month half-life - so a rating from two years ago would count at one-quarter its original weight. The proposal raises practical and fairness concerns: people with a record of bad behavior might improve their scores simply by waiting, while infrequent users may lack enough recent data for reliable estimates. Possible safeguards include Bayesian smoothing or requiring a minimum effective sample size. The discussion also asks whether platforms should show both lifetime and recent-behavior scores, whether a fixed 500-rating window is statistically defensible, and how reputation systems can balance predictive accuracy with a person’s ability to move beyond subjective, outdated assessments.

Read on news.ycombinator.com5 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in Culture

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.