User Poll Ranking Algorithm summary

In fact, this can only be considered a summary, a review of the takeaway. "Ranking algorithm based on user voting" series, Author: Nanyi. This series of articles is indeed beautifully written, layered and thoughtful. Author blog:http://www.ruanyifeng.com/blog/
1. Only the number of votes in the affirmative (1) per unit time is ranked delicious. Implementation: hourly statistics times, high in the front. Advantages: Simple, it does reflect the heat. Cons: not smooth enough; popular content may have been in front. (2) Time decay poll number hack news implementation: P is the number of votes, T is time, G is the attenuation factor. Pros: There are elimination strategy disadvantages: can only handle praise.
2. Pros and cons of the difference (Reddit's hot article) realization: Z for pro-objection, Y for pro > objection. 1:-1 Advantages: Considering the affirmative and opposing behavior, but also based on time decay, disadvantage: basically by the posting time decision, super popular article will be ranked in the front, general popular articles, controversial articles are not very forward.
3. Participation degree, quality, time. StackOverflow implementation: Qviews, Qscore (question score), qanswers (number of answers), Ascores (answer score), Qage (time to issue from question), qupdated (time from last answer) Advantages: According to the actual needs, the combination of quantity, quality and heat. Cons: Ascores is simply added, and it is possible to overwhelm the weight of a particularly good answer.
4. Newton's law of cooling is actually the law of decay by time.
5. The Wilson section (Reddit's review ranking, currently uses this algorithm. Implementation: P represents the "pro-vote ratio" of the sample, n represents the size of the sample, and the table Z1-A/2 shows the z-statistic corresponding to a certain confidence level. (this formula) if n is very small, the lower limit value is significantly less than p. Advantages: It solves the problem that the turnout is too small and the result is not credible. Cons: Leaderboard is always the most votes in the project, new projects or unpopular projects, it is difficult to have a chance.
6. Bayesian averaging (IMBD) implementation: C Average number of votes per item, m overall average, x value per ballot, N turnout merit: introducing a priori probability, each new poll is an adjustment factor that keeps the overall average score close to the actual voting results of the project. Makes the quantity factor average, the influence becomes smaller. (Quantity is attenuated by time.) In front of it, time decay makes the effect of increasing the number of growth smaller) disadvantage: Assume that the user's vote is normally distributed, unable to highlight debate.
The general statement: it is obvious that seemingly simple problems actually have many factors to affect, of course, for their own projects to be dealt with accordingly, different weights and different factors will inevitably lead to the results.