Comments/Ratings for a Single Item

The heart icons look very nice.

But how about discounting the self votes, and sorting by attenuated score? If the self votes are still counted in the raw vote count, and the latter is used as primary sort key, the incentive to favorite all your own games still exists, and nothing has really changed. A number between parentheses hasbeen added, but no one will really care about what that says if it is not used for sorting.

I have now modified the script to calculate an attenuated score for each game. This appears in parentheses, and it is used as a tie-breaker in the sort. The code for calculating this value looks like this:

foreach ($votesbyperson as $key => $val) {
    if ($val > 1)
        $votevalue[$key] = min(1, 60/(50+$val));
    else
        $votevalue[$key] = .5;
}

This attenuates the value of single votes and the value of lots of votes while counting 2-10 votes as 1 vote each.

It looks great, Fergus, and I wouldn't even have needed the legend to figure it out.

I have changed the heart images used on this page. Your own favorites are now indicated by a heart with an arrow through it. An inventor's favorite is indicated by a heart with a star, and this occurs after the name of the inventor who favorited his own game. Positioning it beside the name makes it clearer what this means even if someone ignores the legend, and when a game has two inventors, it lets us know which one favorited the game. For games you did not favorite yourself, an orange heart is used. I tried the one that is red, but it appeared as a small black heart.

Be careful of not making generalities of things which have a low probability.

• How many people have voted for more than 50 or 60 games? If they are 2 or 3, it should not be difficult to send them an e-mail saying that from now only 50 or 60 votes are possible, so they should re-do their votes if they still want to favor some of the games. Their previous list is simple nulled. They should not complain as they old list was actually meaningless and their new one will make some sense.

• Who can seriously convince many friends to come on CVP, register and then favor one precise game (the same for all of them) and then go? I don't believe that. Maybe it will happen for very few just asking 1 friend to do this. But I doubt. And if it happens, who are we to judge? After all, if that friend really likes his frieds's variant, what's the problem? Who are we to say that this is not sincere? This is making a lot of trouble for a very small issue which is probably not occuring.

About self-vote, I take the point that there are other ways to indicate the author's preferred games. If there is this facility, then there is no need to have the self-vote coming in the same basket than other votes. So, the simplest is probably to forbid self-votes.

It seems much easier to just print an asterisk behind the number of people that voted for it. The info naturally belongs to that number: "5 votes, but note that one of those is from the inventor".

I'm now thinking that the value of a self-vote should be influenced by how many self-votes someone has cast compared to how many games he has. If someone favorites all his games, his self-votes should count for nothing. If he favorites only a few, and he has invented many more games, each may count as a full vote. In between, there could be some attenuation for excessive self-voting. It's a matter of deciding what the limits should be and creating a formula.

I think self-voting should not count for anything as far as ranking is concerned. Counting it, even slightly, and with an unrelated penalty, would still encourage undesired behavior for some. It should solely be used for indicating to the user what the inventor considers the best games amongst his own.

Ideally the number of self votes should be a fraction like 25-33% of the inventor's games. But it would probably be also acceptable to ignore games that were not favorited at all, and let the script only consider the set of favorited variants. Perhaps the following should work: there can be two symbols printed with each variant in the favorites list, one meaning "inventor thinks this game is good", the other "inventor thinks this game is bad". If an inventor than self-votes for more than half of his games in the favorites list, the variants he voted for would remain unmarked, and the smaller number he did not vote for get the second marker. That should encourage excessive self-voting.

Indeed casting a single vote is suspect. I already suggested there might be a case for having the weights decrease for very small numbers of votes. Of course it doesn't really help against cheating; inventors could simply ask their friends to cast several votes. I guess no system can be resistent to such 'friendly cheating', for persons that have sufficiently many friends.

I could change the heart color

For the color-blind or for someone using an e-ink device, I should probably use a separate image. The lightbulb would work on the Favorites page.

where would this self-favorited badge go? Just on this page, or everywhere on every index page, or on the game page itself?

The game page already mentions when it is a favorite of its inventor. For the Favorites page, I could change the heart color for a game favorited by its inventor. Adding it to the index page could be relevant if we're looking at the games of a particular inventor, but it probably doesn't need to be included otherwise.

One more thing I think we should attenuate is the worth of a single vote. Some people may have gotten their friends to come here to vote for their game, and these friends may not stick around and take an interest in other games. Voting for more than one game demonstrates an awareness of multiple games, whereas voting for only one does not. Perhaps a single vote could count for half of a regular vote, or even not count at all when calculating the attenuated score.

@Jean-Louis, where would this self-favorited badge go? Just on this page, or everywhere on every index page, or on the game page itself? You could just list such on your About page, but that wouldn't be very visible. I think it would need to be restricted, or we have the same issue as now: self-favoriting all your games doesn't hurt the ranking or visibility, and the reader has to look over a lot of games to notice what's happening and decide they don't care about badges given by a self-congratulatory author.

I discovered the problem with no-information Favorites on this page: their ItemID's are incorrect, generally for containing a hyphen between words where the actual ItemID has just concatenated the words. I remember there was some issue around that, but don't remember the details. I can just remove hyphens in the ItemID field of the Favorites table, but I'm not sure if there's an underlying problem that would need to be fixed to prevent future issues.

(For other editors or me later: I found these by querying Favorites left join Item using(ItemID) where Item.ItemID is null.)

If we don't want self voting, we can just forbid it

I never said anything against self-voting. My issue is with mass self-voting, and forbidding self-voting is not the solution to that. That's why I suggested penalizing excessive self-voting.

But self voting can be useful, because inventors presumably know their own inventions better than anyone else, and can indicate to others what they feel are their best designs that way.

Indeed. That's why I allow it.

But that purpose would be completely defeated if inventors are encouraged to vote for all their inventions. So we should not encourage that by making it a good way to get those closer to the top of the list. A modest amount of self votes is very helpful, voting for all is as useless as voting for none.

Yes, exactly.

There should not be any penalty for the desired behavior.

I'm now thinking that the value of a self-vote should be influenced by how many self-votes someone has cast compared to how many games he has. If someone favorites all his games, his self-votes should count for nothing. If he favorites only a few, and he has invented many more games, each may count as a full vote. In between, there could be some attenuation for excessive self-voting. It's a matter of deciding what the limits should be and creating a formula.

I find Fergus's principle quite interesting. I like these threshold effects that lower the value for those who like everything. Looks fair to me.

Concerning own's vote, I wouldn't be upset if they were simply made impossible. To compensate, as I said earlier, let the author having the possibility of declaring that this variant is among his ones he prefers. If he puts this "badge" on all of his variant, then, he will simply spoil his chance to make a distinction.

To illustrate with my own case. I will ok to say that my vote on Shako, Metamachy, TerachessII is removed because I am the author. But I would put a "badge" on them, and not on Perfect 12, Exchess or Teramachy which are not my preferred game. It is an interesting information I can give to the reader.

I am not sure editors should be treated different than others. I am an editor, and Jean-Louis is not, and yet I see no reason why my opinion would carry higher weight than his.

I also don't think it is a good idea to 'penalize' self-voters by reducing the weight of their vote on other people's games (which is basically an unrelated issue). If we don't want self voting, we can just forbid it; that is more effective than merely discouraging it through some arbitrary penalty. But self voting can be useful, because inventors presumably know their own inventions better than anyone else, and can indicate to others what they feel are their best designs that way. But that purpose would be completely defeated if inventors are encouraged to vote for all their inventions. So we should not encourage that by making it a good way to get those closer to the top of the list. A modest amount of self votes is very helpful, voting for all is as useless as voting for none. There should not be any penalty for the desired behavior.

Anyway, the trial page I made already does the dual scoring. But it always sorts by the weighted total. I guess the easiest way to make this user-selectable is just have two versions of the script, where clicking a link or button would navigate you from one to the other. (It could be the same script with a different CGI argument, which then decides what to use as sort key.) Only slight adaptations would be necessary to alter the weighting formula.

That's the basic idea. And now I will describe some modifications to address other concerns. If we change 50 to 40 for editors, this will give editors 20 unattenuated votes instead of the default 10. If we add another variable into the calculation, such as min(1, 60/(n+m+50)) where m is the number of one's own games someone has favorited, this would cause a vote's worth to diminish at a higher rate for people who favorite their own games. If we wanted to more strictly penalize the mass favoriting of one's own games, we might replace m with something like min(m^2, m!). For values up to 3, m! would be lower, and for higher values, m^2 would be lower.

I'm now caught up on this discussion. I would propose having the script calculate two scores. One would be the raw score it calculates now, and the other would be an attenuated score that factors in how many games someone has marked as a favorite game. The formula I propose using for determining the value of someone's vote in calculating the attenuated score is this: min(1, 60/(n+50)) where n is the number of games someone has favorited. This will create fractional scores, which I'm not bothered with, as this would be a supplement to and not a replacement of the raw score. The page would then give two options for sorting the list. Since there is a difference of 10 between 60 and 50, this would allow someone to favorite up to ten games without any attenuation of his vote in the attenuated score. After ten votes, his votes would slowly reduce in value. At 70 votes, each vote would be worth only one half a vote. At 130, each would be worth one third. At 190, each would be worth a quarter. Continuing to add 60 more votes, 250 would be worth one fifth, 310 one sixth, etc. If we decide we want the value of votes to attenuate at a slower or faster rate, we could use different values than 60 and 50.

The one-man-one-vote system is not so very different from the 1/sqrt(N) weighting. People that have casted 43 votes (close to the maximum of 50 that you propose) already get weight 3. So a weight of 19 does not advantage you as much as it sounds (but still a factor 6). It is mainly those that have casted more than 50 votes, which you don't want to allow at all, that get heavily discounted. With a more typical 10 votes the weight is still only 6.

With a cube root someone casting 48 votes would get weight 2, and those with more would be 50% discarded rather than being forbidden / ignored. While casting a single vote would give a weight of 7, only 3.5 times larger. With 10 votes the weight would be 3, only 50% more than with 48 votes.

The problem is that people giving 270 and 389 votes are already in the database, so what should we do with them? Completely ignoring all their votes seems a bit harsh, requiring them to 'unfavorite' several hundreds of variants a bit inhumane. It is like you say: if someone favorites tons of variants, it doesn't mean much. So it seems reasonable to reduce the weight of those votes, but a bit inflexible to reduce it by 100% always.

Perhaps the weight should be made to saturate at twice what you get for ~50 votes. Or even decrease if you favorite too few variants. That would thwart attempts to 'pump up' a single variant as a favor to a friend by people who otherwise do not care.

A special symbol could be added for variants preferred by its inventor. Like an asterisk behind the number of people that voted for it. This raises the question of how to put bounds on that, preventing inventors to automatically favor all their variants. Writing such a symbol is an all-or-nothing action, so you could do little else than omitting the symbol on all variants for an author who favorited, say, more than 25% of his submissions. This would then raise the technical problem that you would have to figure out how many games each author submitted, as he could have games that no one favorited. So going through the favorites list would not be enough, you would have to go through the entire database.

An alternative would be to allow one plus the number of his favorited variants divided by 3 (rounded down). Then inventors that never got any votes from others can only vote for one of their games (to vote for two you would need three favorited games). With 10 games favorited by others you could either vote for 4 of those, or 6 different ones.

I am not convinced by the fairness of one-man-one-vote. A variant that gets a vote from someone that voted 50 times can very well, even in the eyes of that person, rank below 49 other variants. While getting the vote from a person that only casted 10 votes means he thinks it belongs to the 10 best. So why should they get equal reward?

Then, I'm not sure it is fair. Someone who is not fan of CV, or who just wants to please his friend who has made Turlutu chess, will vote for Turlutu chess and his voice will count x times mine who is enthusiast of CV and wanted to reward those I like or admire.

I prefer the system of 1 man 1 vote. Simpler and immediatly meaningful.

I would admit that self-votes are not accounted (but then, it would be nice to have a special sticker indicating "this variant is a preferred one of its author", because I wish to indicate which ones of my creations I prefer).

To avoid people having tons of preferred CV, which is like saying nothing as a matter of fact, I would understand that each user is limited to a maximum, maybe 50. With the possibility of removing a "like" of course, in order to favorise new ones.

Indeed, that is correct. If you have favorited 4 times as many games, your votes get 2 times lower weight. To avoid fractions, the person that casted the most votes is given weight 1. Currently that is someone who casted slightly less than 400 votes. That means that people that casted 100 - 400 all get weight 1, people that casted 44 - 100 votes get weight 2, casting 25 - 44 votes you get weight 3. If someone casted only a single vote that vote gets weight 19. So the weight increases with fewer votes, but it is not that everyone gets 400 votes and can give all of those to a single variant.

The general formula here is weight = sqrt(max_given_votes / given_votes), rounded down to an integer. Max_given_votes currently is 389, and sqrt(389) = 19.72 -> 19. Giving 2 votes the weight already drops to 13.93 -> 13, and for 3 votes to 11.39 -> 11. So another way of looking at this is that you get more votes to give if you distribute those over more variants.

This is still open for discussion. (E.g. we could also use a cube root instead of a square root, in which case the weight for a single vote would go up to 7, instead of 19.) But considering the large score the top contenders get this way, allowing the weight of a single vote to run up to 19 does not seem too disrupting.

I'm not sure to well understand. With the weighting factor, someone who has favorised many games has a voice which counts less than someone who has favorised only one game. Is that correct or wrong?

I am not sure what you mean.

I now modified it to discard inventor votes in the calculation of the weighted total. Self-votes are still counted for determinig the weight a voter should get, though. (But because of rounding having a few extra self-votes usually would not decrease your weight for the other votes.) At the moment it displays the weighted total including the self votes, and the number of people voting for it (including self votes) in parentheses, just to judge how large the effect of these refinements is.

An interesting consequence is that there now also appear some games with a zero total at the bottom, where only the inventor favorited those. By comparing with the total including self votes it can be seen that weights run up to 19! (Presumably for persons that only favorited a single game.)

That is cool, but maybe you could include a display of the rank!

I had a go at improving the ranking of favorites by weighting the votes, approximately as 1 over the square root of the number of votes cast by the user, normalized to the most-voting user getting weight 1, and rounded down for the others. This mainly as an exercise for testing whether I understand enough of how the PHP scripts work here to implement such a change. There is no correction for self-voting yet.

The modified page is here. It displays both the total of the weighted votes (on which the list is sorted) and the number of people voting for it (the old sort key, which is now used as primary tie breaker).

Since the "127 favorites" from my 2018 comment on this thread came up, let me point out the reporting page that generates the table on demand. That user, right now, is up to 147 favorites, but has been overtaken by two users, with 187 and 389 favorites. Another thing to note is there are ~170 users who have favorited something. I'd be interested in seeing something broken down by self-favorites.

Well, it doesn't really matter what the maximum is. The score calculation can determine it. It could just scale up the weights of users that favored less than N-squared times fewer games by a factor N.