And your opinion on film search is taken into account?

Watching one of the films on film screening, I discovered how suddenly one day his rating dropped sharply. Intentionally I will not specify the names of movies and nicknames of users, so as not to cause unintended consequences. I tried to figure out the reason, and found the following: it turns out that the rating is not always calculated in the same way. All users of the film search are divided into 2 grades: the ratings of which are taken into account in the rating, and accordingly are not taken into account. After examining the statistics, profiles and rating history of a multitude of users, and having written a small program to analyze the distribution of ratings, I found out how this happens.

At first, when there are very few votes, the rating is not shown at all, but you can still see it in the profile of the director or actors. Then, when the number of votes is still small, the rating is calculated simply as an arithmetic average. And only then, when the film gains enough votes (usually 50-100), the basic formula comes into play. In the help of film search about the calculation of the top 250 is written:

Based on user ratings, KinoPoisk generates a rating of the Best Top 250 films. Some users want to influence the rating of the best films by registering several accounts and / or exposing films to only a few or dozens. Therefore, only regular users who honestly evaluate films and their impression of them take part in the calculation of this rating.

About the ratings on the pages of movies there is nothing said, but obviously a similar filter is used not only when calculating the top 250, but also for all other films. Having figured it out, I naturally became interested in how they determine whose estimates to take into account, and whose not. The administration does not disclose this information, which is understandable: the main purpose of introducing such a filter was to weed out unscrupulous users who deliberately underestimated the ratings of some films, and overestimated those of others.

At the film search, the rating is shown with an accuracy of up to thousandths, so for any film where there are less than 100 votes, it is easy to determine how many votes were taken into account. The number of options, when the division of the integer sum of estimates will be exactly equal to the given number, taking into account the rounding, is not so great, usually 1-3 options. You just need to calculate the amount for each possible number of ratings from 1 to 100, round off, divide back and compare with the desired rating.
')
Thus, I determined that out of almost 100 ratings given to the film I was interested in, only 30 were counted. ratings) for the film on the timeline:



It is clear that most of the ratings were given exactly by habrap users after the post was published. It turns out that 70% of these estimates are not taken into account. Judging by my friends, people related to IT are rarely active participants in film screening, and although many of them have been registered there for many years, during this time they gave very few ratings. So for this film, judging by the reviews on the rutracker, a similar picture was observed:

The first time I say thanks to someone on the tracker.

Finally, something that really interested me is enough to run into distribution every day and watch for updates!

There were quite a few such messages, including from people who for many years wrote the first time on the rutreker. The film is so unique that because of it, people first rated it in a movie search. The picture at the beginning of the article is a graphical representation of this particular case: the statistics of assessments, consisting of one assessment. And many just signed up to put this single assessment. And their opinion is not taken into account? I think this is unfair.

Determining whether your estimates are counted is simple. It is necessary to find a film with the number of estimates of about 50-100. If the number of votes is too large, the change in rating from your rating will simply not be noticeable, and if it is too small, it will fall into the category of films, where all ratings are still taken into account. Then you have to vote. If the rating changes (and it changes immediately after the page is updated), then your opinion is taken into account. To be sure, try several films. But then remove the estimates, because we, as honest users, should vote only for those films that were watched.

I decided to find out what the selection criteria were. All estimates and their exact amount are known to us. It remains to find out which of the estimates are taken into account. For some time I tried to manually select the options so that the sum of all the estimates gave the required number, but it took a lot of time. Therefore, I wrote a small program that, using the recursion method, went through all the possible options for distributing exactly how many ten, nine, etc. of all ratings taken into account in the final rating.

Here is its main loop:

private void count_cases(int iter, int prev_sum, int max_rest) { //  .    max_rest = max_rest - max[iter] * iter; //   ,      //   -     int cmin = (sum - prev_sum - max_rest + iter - 1) / iter; if (cmin < min[iter]) cmin = min[iter]; //   ,      //         int cmax = (sum - prev_sum) / iter; if (cmax > max[iter]) cmax = max[iter]; //     for (int i = cmin; i <= cmax; i++) //     count_cases(iter - 1, prev_sum + i*iter , max_rest); } 

As a result, despite the 10-nested recursion, the program quickly gives all possible options, for example:


Now it remains only to look at user profiles and suggest why the estimates of each particular one may not be taken into account. The criteria may be the following: small experience, few evaluations, or, as we learned before, the assessments themselves are such that film search does not trust them.

Specific figures have not yet been accurately determined, as there are many options and statistics are not enough, but you can already say that you need to have experience for several months, and put a few hundred estimates to take your opinion into account. And put them not in one day, but for quite a long time. All this is rather trivial. But studying the history of ratings, I found a lot of interesting things. Although most of them look pretty standard:

Or estimates are shifted down or up slightly:

But there are also special cases, for example, the typical categorical one and tens. It even happens that there are even 1000 ratings, some 10s. I must say, this is quite common, that is, if you look at 10s from a regular film, you will most likely find just such a picture:

There are also people with a "dubious past." In the beginning, they only ranked 10k, and then their ratings acquired a normal distribution:

In the end, it still turns out that out of 1000 films that people watched, half are masterpieces. It is clear that film search considers such an opinion inadequate. Such users, in order to “improve”, need only to press the “delete all ratings” button, or carefully review all their tens and units, and arrange them differentially.
There are also people who do not like anything at all:


I have seen many more interesting graphs; in fact, it is possible to study the life lines of film search users. Analyzing them, selecting the possible options for the distribution of estimates and comparing all these data among themselves in each case, I was able to determine exactly which estimates are taken into account, which I wrote about above. Of course, I could be wrong, because there may be other, very non-standard selection criteria, which we can only guess. I will not describe the process in detail right now, I can write a whole article about it, but in the end we can say this:

On film search, if you are not an active user, your ratings will not affect the rating of the film at all, even if you like it very much. But this does not mean that you put them in vain. They are taken into account in rare films, where only 5-10 votes. They are taken into account when searching for friends by interests and selection of recommendations, what to see. And of course, if you want your ratings to be taken into account, give more ratings, and eventually, over time, you will move into the category of users whose opinions are considered.