What is a fair formula to use to calculate the true likes (likes) for an answer on Toaster?

Question

Ratings

The fact is that the dynamics of evaluating answers is initially not fair. Because the answers appear with a time difference . Because of this, the first answers are more likely to pick up random likes, although they may not be complete and accurate information, and sometimes even completely wrong. Thus, a quick response (even if not quite to the point) attracts more likes, all other things being equal.
After all, you can’t minus on the Toaster. So the response time is related to the number of likes anyway.
Later answers have to compete with the first, already overgrown with likes, answers. Of course, it is possible to rise to a suitable answer, but only if the time difference is not large. For clarity, open a year-old question without a solution, even with a lot of subscriptions, and try to give a really good thoughtful answer compared to the available ones. Let's say the first answer has 3 likes. Your answer, God forbid, will have 1 - and that's it. Conformity and laziness
aggravate the situation. That is, even if a person enters a popular question that already has all 10 possible answers (and no more is expected), then he will read the first two answers, which have 17 and 14 likes, respectively, and other answers with 5 or less likes cannot be read. will (probably). Moreover, the first two can even line, which formally means that they are better than the others, although the user did not even compare.
The question is, what formula can be used to calculate real sympathyto the question, given the above factors? That is, the number of likes for each answer is known - these are integers, and the exact time of the question and each answer is known (timestamp accurate to the second). Well, the current time is also known, of course. In short, the conditions of the problem are the Toaster. Based on these data, you need to somehow more accurately measure sympathy. It can be expressed as a real number.
What formula could it be? How to get it out?

Nuances

- Пользователей мы не считаем абсолютно разумными. Всегда есть шанс, что лайк достанется чуть ли ни случайно. Но статистически всё же проявляется разумная симпатия. Т.е. люди не полностью рандомны, иначе формула достаточно простая будет.
- В симпатию входит не только оценка правильности ответа, а симпатия в широком смысле, то есть это и эмоции по стилю, оформлению и всякое такое. Нам не нужно знать, что туда входит, нам нужно лишь понять реальную (объективную) оценку по формуле, а не номинальную на основе счетчика нажатий на кнопку нравится, что бы сама кнопка ни значила.
- Самое важное: мы исключаем читерство. Ведь в теории накрутить лайки очень легко. Но это бред. Так что исходим из того, что никто в здравом уме этим заниматься не будет. Так что имеющиеся лайки принимаем за чистую монету.

Answer 1

https://www.reddit.com/r/OutOfTheLoop/comments/4dl...
https://medium.com/hacking-and-gonzo/how-reddit-ra...
There is no such thing as justice
A without clear definitions this word most often hides the deception of the majority in favor of a certain group of people
. But max/min and normalization (not sure about this term) really exist, are predictable and can be tested
google: reddit ranking algorithm

Answer 2

Beg your pardon, captain, but do you write an answer to get likes? I'm here - no, and I usually pooh my answer rose or did not rise. It's nice, of course, when you see what people like, but when they don't like it, it doesn't bother me at all.

Answer 3

First of all, it will be necessary to collect statistics on the time of appearance of each like from the moment of the answer, but as far as I know, the time of a like cannot be viewed here. Although if you ask the admins, they can give this data. If there is a time for the appearance of likes, then proceeding from this, dance further, for example, assigning coefficients to the likes by which they should be multiplied: that is, if let's say statistics showed us that during the first hour answers receive more likes, for example, 2 times than for all the rest of the time, then each like is counted as 1, while all the likes after an hour are counted as 1 * 2 to equalize unequal conditions. But this is a trivial example for demonstrating the idea of ​​equalizing likes, in fact, more multipliers are needed and various time intervals should be carefully examined. In addition, the day of the week and time are of great importance. when the question was asked. On the same weekend rotten.
Regarding liking for style, etc., this is also very difficult, because you first need to analyze everything and correctly classify all styles, which is not so easy, but recognizing these styles is even more difficult. Also, the likelihood of likes increases if the question has a lot of views. And the number of views depends primarily on the tag to which the question belongs, but no less significant point is the title of the question - the more clickbait, "beginner" and "watery" it is (like where to start learning x), the more clicks on it , because clickbait gives interest, and the rest simply increases the number of people able to give an answer. Well, those who want to joke too.
This is just the first thing that came to mind, in fact, a lot of factors that you cannot fit into one formula in my opinion.

Answer 4

I'll try to describe it in the most understandable way:
1. Without likes - everyone competes, because. there are still few of them, but with unequal probability: the later the answer is received, the lower it is in the list and it has less chance (that it will be read).
2. After any like - this answer rises to the very top and begins to compete with the next 2 standing below it (by posting time).
3. As soon as the cluster is assembled (1-3 answers with a strong margin from the others) - the rest - almost no chance.
Then:
1. If the person who asked the question understands the topic, he will read everything and choose 1 (rarely 2) answer FROM ALL!
2. If he does not understand, then the "pop-up" cluster (with the maximum number of likes) becomes the solution.
Based on this algorithm (process), you need to takethe average time spent reading (T1) after opening the page with a question (and leaving the page without a like) and the average time spent reading before setting the first 2 likes (also after opening the page with a question) by different people (T2).
Then take the average of these 2 averages: (T1+T2)/2
And get the interval time peak when reading the first two answers.
Next - by the number of characters (according to the leading answers), we calculate the average reading speed and, approximating, calculate ("level") the time to like for each answer (from the first and those below).
Now we have 2 coefficients:
T - reading time with an adequate like for the 1st and 2nd answers together (the average was there)
S - reading speed (number of characters per unit of time) with an adequate like
When liking the downstream - we multiply by the time to normalize to the peak of an adequate like.
Further from the peak - less score.
Closer - more score.
That's the whole formula.

Answer 5

It’s easier to just mix up the answers and not show likes until the question is closed, as soon as the question is closed, then the true answer will be visible
Or until the user votes, but if he is not logged in, then you can show everything as it is
Well and if you approach it mathematically, then you first need to create a map of views and clicks as in metrics, drive it into a neural network and display the percentage of accessibility of each answer to the question, further calculating the difference in percentage and the number of likes . Or somehow so factor about in a lot and this answer can you
to crawl up despite the fact that it is out of the

Answer 6

The question is, what formula can be used to calculate the real sympathy for the issue, given the above factors?

No.
If you want fair, balanced assessments, then replace the user with a predictable robot.