Ranking algorithm based on user voting (vi) Bayes average

The previous article introduced the "Wilson Interval", which solves the problem of too few votes and the results are not credible.

For example, if only 2 people voted, the lower limit of the "Wilson interval" would sharply lower the percentage of votes. This ensures the credibility of the rankings, but also brings another problem: the top of the list is always the most votes of the items, new projects or unpopular projects, it is difficult to have a chance, the ranking may be long after.

Taking IMDB as an example, it is the world's largest film database, and viewers can vote for each film with a minimum of 1 points and a maximum of 10 points.

The system calculates the average score of each movie according to the voting results. Then, based on the average score, the top 250 most popular movies are expelled.

Here's the question: is the average score between popular and unpopular movies really comparable? For example, a Hollywood blockbuster has 10,000 viewers voting, and a small-cost piece of literature only 100 viewers voted. How do you compare the results of these two polls? If the "Wilson interval" is used, the latter's score will be significantly lower, is it fair to deal with, and can it reflect their true quality?

A reasonable idea is that if you want to compare the quality of the two films, you should at least invite the same number of viewers to watch and score. Since the number of art films is small, we should try to increase the audience for it.

At the bottom of the ranking page, IMDB gives its computational method.

Url:http://pic1.bianceng.cn/programming/sjjg/201410/46061.htm

-WR, weighted score (weighted rating).

-R, the movie's user voted average score (Rating).

-V, the number of votes in the film (votes).

-M, the minimum number of votes for the top 250 films (now 3000).

-C, the average score for all films (now 6.9).