Random thoughts on the proposed changes:

• Doing the default-3-rating thingy is going to put a lot of data into the db. That's probably not a good idea. If you have to pad your db with lots of default data, you're probably doing something wrong.
• I suspect that in practice, having every user rate every comment a "3" or "neutral" will result in so many ratings that zeroing out spam (or anything else) will be effectively impossible.
• I'm also a little skeptical about your assertion that the problem is a lack of rating. This comes across as "the problem is that the users aren't working hard enough." Glass houses, and all that.

I actually think the problem should be stated quite differently from what I see here, but in the interest of keeping things on-topic, I'll keep my trap shut and just point vaguely at my preliminary thoughts on the matter until such time as I have some working code to illustrate any hairbrained scheme I might come up with once I've got a better handle on things.

As I'm reading through comments, I don't mind rating them along the way (though it might be nice if it were one click instead of two, say, with radio buttons). What irritates me is that at the end of my rating spree, I have to hit some "rate all" button and wait for the (possibly slow) other end to present me with another page.

Will your default ratings really work, if no one ever hits "rate all"?

I'm not quite clear on what automatic rating means here. Does it automatically rate whenever you move to another page? That would be good.

Anyway, let's solve your problem of when to hide comments. We can sort of map a binary distribution onto hide/non-hide ratings, as follows: For a comment with n ratings, let h be the number of hidden ratings. We simply ignore the actual value of all other ratings. Now we've got a binomial distribution and the equations are easy.

First, the mean deviation: we need a "probability" for hidden ratings, for now we'll just let that be p=1/4. sigma(n) = sqrt( (p) (1 - p) n ) => sigma(n) = sqrt( 3 * n / 16 ).

Now take the difference between h and the expected number of hidden comments p*n, so the difference is: h - p*n. Divide by sigma(n) to get the number of deviations: (h - p*n) / sqrt( 3 * n / 16).

Okay, so given a comment with 16 ratings having 3 "hide" ratings, do we hide it? Well, plugging in the numbers we get: (3 - (.25 * 16)) / sqrt( 3 * 16 / 16) => -1 / sqrt(3) => -.58. A negative value means there are fewer hide ratings than expected, so don't hide this one.

Given a comment with 64 ratings and 20 "hide" ratings, do we hide it? That is (20 - 16) / sqrt(12) or 1.15 deviations.

Now you get to choose where to set the allowed deviation for hidden comments, and +1 deviation might be good.

You can also choose the value for p.

Another side effect of this method is that below a certain number of ratings, it is impossible for comments to be hidden. Say a comment has 3 ratings and they all say "hide". Well, that is a deviation of only .97, which is less than 1. So the deviation method gives much different results for small samples than the naive "averaging" method -- one could say that it will not give you false positive where there are too few samples statistically.

Of course, one could just not take action against a comment with less than some number of ratings. Use of the mean deviation tell you why this is the correct thing to do.

The simple case, which I mentioned over on k5, is one in which you assume that 50% of readers change the default. It then becomes impossible for a post to be made invisible; 50% don't modify the default and assign a '2', 50% *do* modify the default and, even if they all assign a '0', you end up with a '1' rating.

You could argue that posts deserving of zeroes are, or should be, so egregious as to result in a higher-than-normal rating percentage. But that leaves the quandary of the normal post: *if* you compress the scale, and if you have a default, a 50% moderation rate would give an effective scale of 1.5-2.5. Values would cluster around the default.

If you wanted to take it a step further, make everyone a trusted user in their own diary, and let them delegate trust to others in that domain.  Actually, the diary owner should always be allowed to zero posts in their diary, since it's their domain.

The problem is that you are going to suffer the reverse effect; positive ratings are going to be buried in the noise of neutral ratings.

Scoop is not very good at figuring out which comments I've read and which I haven't. Specifically, in minimal display mode, I'm considered to have read comments when I've only seen their subjects.

I'd suggest marking a comment as read only after the user rated it. That way, any comment I haven't rated always shows up as [new], whether I've seen it or not. If I'm following a discussion, I'm encouraged to rate every comment I read, just to get it off my radar.

For example: I like the ratings tps12 and TheophileEscargot are casting, so I want my initial ratings on comments to be the average of the votes that those two persons are casting. And of course I want to be able to fix the comment ratings from the initial to the rating I am wishing to.

But (as briefly discussed in #scoop) I think that not having a "none" rating available, and having all rating selectors set to 3 by default would work. That way if somebody wants to rate any comment, they rate them all; if they don't want to rate at all, they don't provide too much inertia.

Personally, if I rate I prefer to rate most or all of the comments in a story but setting most of them to 3 is kind of annoying, since I only think of going to the rating box if I want to rate something other than 3. If a comment is average, it doesn't inspire me to choose a rating - so the default rating should be "average".

--
Discuss the art and craft of writing

I don't think I much like the idea of auto-rating things seen.

• I think reading should be passive: until you actually hit a rate button, you're not intending to rate anything: that a default rating would occur would seem surprising.
• Just because a comment appears on a page does not mean you have read it.
• For any number of reasons, you might abstain from rating a particular comment
• This would add too much intertia, I think.

In my opinion, a better solution, would be to establish certain range limits of ratings a comment could have that depend on how many people have rated it.

For example: a comment with 1 rater could rate between a range of 1.5 and 3.5, ...

 switch(number_raters) {
    case 0:  rating = 3;
    case 1:  rating = min(3.5, max(1.5, average));
    case 2:  rating = min(4.5, max(1.0, average));
    case 3:  rating = max(0, min(5, average));
 }

Well, that would add a tiny bit of inertia.