aniDB Hint Reconfiguration [closed]
Moderator: AniDB
-
- Posts: 144
- Joined: Fri Mar 24, 2006 6:29 am
- Location: Israel
aniDB Hint Reconfiguration [closed]
Hi, sorry for not using the tracker but I didn't get the validation e-mail for it.
I was hoping that you'd allow for an option in the aniDB Hint that would allow users to assign their own values to gaps between votes. e.g.:
0 = 100
1 = 40
2 = 5
3 = -20
4 = -60
5 = -100
etc...
Also, for some reason the avg. adj. doesn't seem to be able to change to any figure other than 5, must be a bug or something...
I was hoping that you'd allow for an option in the aniDB Hint that would allow users to assign their own values to gaps between votes. e.g.:
0 = 100
1 = 40
2 = 5
3 = -20
4 = -60
5 = -100
etc...
Also, for some reason the avg. adj. doesn't seem to be able to change to any figure other than 5, must be a bug or something...
I am getting close to the end of a big project (the customer is supposed to have it by Monday), so maybe I will be able to start looking at the hint again... Well maybe not, I just got a call, after typing that saying we might get a big order in next week. We'll see...
Anyway, although the option to specify the gaps between votes sounds interesting, my guess is that it would not be that useful anyway and would probably increase the load because people would keep playing with the numbers and running it again. I think the hint needs a whole new approach, the existing mechanism has been tweaked a lot, but even with different tweaking the results are not that much different. I was working on a whole new scheme, but got too busy to continue... So, modifiable gaps I don't think should be done.
As far as the avg. adj., it appears to be working for me. If I remember correctly, it takes whole numbers from 0 to 99999, anything else will revert to the default (5).
Anyway, although the option to specify the gaps between votes sounds interesting, my guess is that it would not be that useful anyway and would probably increase the load because people would keep playing with the numbers and running it again. I think the hint needs a whole new approach, the existing mechanism has been tweaked a lot, but even with different tweaking the results are not that much different. I was working on a whole new scheme, but got too busy to continue... So, modifiable gaps I don't think should be done.
As far as the avg. adj., it appears to be working for me. If I remember correctly, it takes whole numbers from 0 to 99999, anything else will revert to the default (5).
-
- Posts: 144
- Joined: Fri Mar 24, 2006 6:29 am
- Location: Israel
The current scale seems reasonable.
It certainly won't improve recommendations if the voting between two users matches otherwise quite well, but just one or two mismatched ratings causes such a huge penalty that they won't affect each others recommendations...
It is necessary to give some leeway for inaccurate ratings - the amount of which, I'm sure, is quite numerous.
There are plenty of things that can cause users to make completely inaccurate ratings. For example, the so-called "bad days" that most people have, when they simply end up hating something because of some abritory minor fault. ...or voting for something that the person watched so long time ago that he can't remember it well enough to accurately rate it. ...or voting for someting that he didn't in fact ever see. (Yes, that happens: "Oh, I think I saw that one a decade ago, and it utterly sucked. I'll rate it... eh, 2 is fine.")
It certainly won't improve recommendations if the voting between two users matches otherwise quite well, but just one or two mismatched ratings causes such a huge penalty that they won't affect each others recommendations...
It is necessary to give some leeway for inaccurate ratings - the amount of which, I'm sure, is quite numerous.
There are plenty of things that can cause users to make completely inaccurate ratings. For example, the so-called "bad days" that most people have, when they simply end up hating something because of some abritory minor fault. ...or voting for something that the person watched so long time ago that he can't remember it well enough to accurately rate it. ...or voting for someting that he didn't in fact ever see. (Yes, that happens: "Oh, I think I saw that one a decade ago, and it utterly sucked. I'll rate it... eh, 2 is fine.")
Keep in mind that most users have a very limited range of votes which they actually use. I.e. only votes from 6 to 9 or 90+% between 7 and 8.
That means that even though a difference of two vote grades seems to be only 20%, in reality it's probably more like 80-90% difference.
So if you reduce the penalty, you're likely to get _less_ accurate recommendations.
BYe!
EXP
That means that even though a difference of two vote grades seems to be only 20%, in reality it's probably more like 80-90% difference.
So if you reduce the penalty, you're likely to get _less_ accurate recommendations.
BYe!
EXP
-
- Posts: 158
- Joined: Sun Jan 09, 2005 10:54 am
- Location: Germany
A reason for that would be, that you normally only watch things which you really like. That's why it's a bit normal to have votes which are 7+.Andemon wrote:Oh, right. Those people. :-7
I had completely forgotten how common such ratings are here. I still haven't figured out why someone would only use two or three ratings when you have the whole scale available.
-
- Posts: 144
- Joined: Fri Mar 24, 2006 6:29 am
- Location: Israel
@Andemon and exp: Well I don't know about other users, but I make it a point to vote for animes as accurately as possible. If after I voted for an anime I think I made a slight mistake for whatever reason, I go back and fix it, and I don't vote for animes I've seen long before discovering aniDB because of the high chance that I forgot their true quality. The range of my votes is all the way between 1 and 9, so understand that I take it very seriously and I really am angry at those people who don't use the voting scale properly.
-
- Posts: 158
- Joined: Sun Jan 09, 2005 10:54 am
- Location: Germany
-
- Posts: 144
- Joined: Fri Mar 24, 2006 6:29 am
- Location: Israel