Forum - View topicBar graph visual ratios
|
Goto page Previous 1, 2, 3 Next |
| Author | Message | |||
|---|---|---|---|---|
Dargonxtc
 Posts: 4463 Location: Nc5xd7+ スターダストの海洋 |
|
|||
|
abunai:
Out of those two options, as a user, I would rather the obvious spammers be deleted, than for the numbers to be taken down. It should help out the visibility of the graph as well, but still take into account the borderline spammers. This way the number of votes would actually mean something. My that really |
||||
|
||||
|
abunai
Old Regular
 Posts: 5463 Location: 露命 |
|
|||
|
Don't worry about it. I know, that's easy to say, but seriously, don't.
As I understand the methodology involved, nobody has a "perfect 100%" rating (even Dan). The system is set up to compensate for skewed ratings, for whatever reason. Let's take the example you give: lack of experience with manga causes you to give more high-end ratings. Why, by the way? Are you only reading the good stuff? If so, how do you recognize the good from the bad? An argument could be made that the heuristics should lower your rating, if you've only got high-end ratings (even if made in good faith), because you might lack experience enough with the bad stuff to form a balanced opinion. As an admittedly inexperienced user, you will naturally have a limited spectrum of votes for the heuristics to judge you. As time goes by (and assuming you give out votes in a balanced fashion), your rating ought to stabilize. But in the final analysis... what does it matter? All you can do, all any of us can do, is submit our fair, honest opinions to the database. Keep doing that, and there won't be a problem. As for those who don't... well... that's what the heuristics are for. By the way, I don't know whether manga and anime are evaluated separately in the database. If they're not, then your limited sample in the manga section is going to be far outweighed by your contributions to the anime section, and the problem solves itself. I suspect it's all one big database, with manga and anime merely being classes of records. But that's Dan's playground, you'll have to hope he reveals more on the subject. - abunai |
||||
|
||||
|
Dan42
Chief Encyclopedist
 Posts: 3791 Location: Montreal |
|
|||
|
The problem is: when do you decide if a certain user is "undoubtedly" or "obviously" spamming? Maybe he just has peculiar tastes? Or just tastes that go counter to whoever is doing the judging? I am not willing to step on the slippery slope of letting humans decide if a particular person has "valid" ratings or opinions.
abunai has a good point about misleading numbers. I could change the graph to exclude users who are below a certain trustability, say, 5% (most spammers appear to be in the 0-5% range) The filter doesn't look for certain characteristics to then flag people as spammer. There is no "line to cross"; it's a trustability continuum, from 0% to 100%.
BTW, manga and anime are not considered separately when calculating a user's trustability. Ultimately, you have to remember that the purpose of the bar graph is to be fair to the people who view the page, not to those who vote. |
||||
|
||||
Tempest
 I Run this place.ANN Publisher  Posts: 10455 Location: Do not message me for support. |
|
|||
When 5+ different user accounts vote identically on an anime within seconds of each other from the same IP, and then do so again on another show, and again on another and... 
-t |
||||
|
||||
|
Dargonxtc
Posts: 4463 Location: Nc5xd7+ スターダストの海洋 |
|
|||
Hmmm. Good point. I wasn't really thinking of it that way. But it does makes sense to me. And the more I think about it, it is actually equally fair to the voters as well. Where as veteran "fair" voters naturally receive slightly more pull for their vote than do the novice "fair" voters. Even if they both rate the ones they have both seen with the exact same ratings.
This is good to hear. Could I also infer from this statement that just like it is possible for ones trustability to go down, that it is possible for ones trustability to also go up? And one more silly question to anyone who can answer it. What would happen if a person who has seen say 200 shows or books, but rated every single one of them a 5? |
||||
|
||||
Vortextk
 Posts: 892 Location: Orlando, Fl |
|
|||
|
Isn't it just the same thing that's being discussed? The only reason that 1 and 10 might be very common is for simple reasons. A person wants an anime to go up on the top lists, or just loves it and is immature about voting. Of course the same applies for the negative side of that, giving it a 1 to drop it or because you hate it and want to spam it. You can also say popular shows like Haruhi get spammed with low scores(as well as high ones I'm sure too) just because it IS popular. People are free to have their opinions, but I really doubt the worth of votes when you see, for example, 3 bad, 5 awful and 54 worst ever.
I would imagine their worth would be just as low as someone who voted 200 1's or 10's, it's just probably more uncommon to spam 5. |
||||
|
||||
rti9
 Posts: 1241 |
|
|||
I give my total support to Dan42, manual moderation is nothing but a pain in the ass.
If the system could identify these independently then it would be great. The thing is to not let human judgement interfere. The rating system is fine. The team that developed it made an excellent job in my humble opinion. Despite all spamming, the top rated lists all seem to be "healthy" (meaning nothing absurd). I would be worried if industry-related users were using it to try to boost sales by interfering in the rating system, but I really doubt that is happening. And even if they were, then it would just prove that the algorithm works. A question to add to Dargonxtc's that might help explain the bar graphs: does a vote given by someone who selects "seen all" and a vote given by someone who selects "seen some" have the same value? |
||||
|
||||
|
abunai
Old Regular
Posts: 5463 Location: 露命 |
|
|||
As a forum moderator, I give you my unconditional ageement on that.
Good question. I don't know how Dan has implemented it, but I'd guess that they do have the same value. The idea of the rating system, after all, is not to produce fixed-in-stone evaluations ("Sorry, you already rated Medabots Excellent, and we're not going to let you change that to Bad, neener-neener!") but rather to reflect the fluidity of viewer opinions. Over time, "hit" shows tend to stabilize at a slightly different level, for instance, as people change their votes (usually downgrading from the Masterpiece ranking they'd originally given, so far as I can tell). An example: many anime that are new in the Encyclopedia, will gather a slew of initial high or low votes, often before they've even aired in Japan. In the first place, these reflect the viewer expectation level, so they are not really "wrong" as such. In the second place, they tend to alter as the people who made the votes (usually the ones concerned enough to actually see the anime) alter them to reflect the actual impression they got from seeing it. And again, these votes are spammish in character, and tend to lower the trustability index of the users in question while they persist, so the heuristics catch this and compensate for it. Someone asked, I think, whether the trustability index of a user can change -- the answer to this, as I know from asking Dan myself, is yes. Apparently the index of a user is recalculated (I'm guessing once a day, as part of regular database maintenance procedures, but it could be less often). I don't know the algorithm whereby Dan calculates trustability, but here's how I would do it: I'd create a bell-curve with a peak centered slightly above average (this is a site for anime lovers, and a reasonable median vote would probably be a vote of Good, so let's be nice an say that that is the norm). Then I'd periodically match each user's actual vote distribution to this bell-curve and calculate how much he deviates from it. This would give me a trustability index that would correct for his inherent bias (which everybody, including myself, has, so there's no onus to the word here). I'm sure Dan's system is somewhat more complex, but that's a stab in the dark at how it might work. - abunai Last edited by abunai on Thu Dec 20, 2007 9:42 am; edited 1 time in total |
||||
|
||||
|
Tempest
I Run this place. ANN Publisher Posts: 10455 Location: Do not message me for support. |
|
|||
Agreed. -t |
||||
|
||||
|
woelfie
Encyclopedia Editor
 Posts: 380 Location: Belgium |
|
|||
Suppose I've seen only part (let's say about 50) of Pokemon's 276 episodes, and I've come to the conclusion that it's really not worth watching one more of them. Obviously, I can't say "seen all". Would that make my opinion less valid that that from the few dozens of people who had the courage to wrestle through the whole bunch, well knowing that they will not alter their opinion (be it good or bad) of the series ? (please don't answer this rhetoric question) |
||||
|
||||
selenta
 Subscriber Posts: 1774 Location: Seattle, WA |
|
|||
I'm pretty sure Dan has said that 'seen all' and 'seen some' are valued the same, as they should be. The belief is that the rating reflects the user's current opinion of the show, and as they watch more or finish it, they will keep it updated. It's too much work with little to no meaningful importance to keep track of how much of a series each user has seen to influence the value of their vote, absolutely unnecessary. So yes, if you've seen enough of a show that you are comfortable rating it, then the 'value' of your vote is just the same as anyone else's. |
||||
|
||||
|
Redbeard 101
Oscar the Grouch
Forums Superstar  Posts: 16963 |
|
|||
|
Here's a question. It may be totally retarded but as they say there's never such a thing as a dumb question.....right?
My question (if Dan can even answer this without giving insider info) is once the encyclopedia is done with all the genre upgrades and search options etc will that effect how a person's ratings are judged? What I'm thinking is some people are more favorable or critical towards different genres. A person may love comedies for example and not care so much for science fiction. So their ratings may be a lot more favorable for comedy titles as opposed to science fiction titles. So does this system that determines a spam rater as opposed to an honest rating consider this option of the genre of the title being rated? |
||||
|
||||
|
Dan42
Chief Encyclopedist
Posts: 3791 Location: Montreal |
|
|||
They are worth the same but I would like to change that. Or at least reflect this datum in the graph. In the most complex case, imagine that the graph would become 3-D, with a "number of episodes watched" axis. Then you could see how the opinion distribution changes as people watch more episodes. e.g. people tend to rate "masterpiece" after episode 1, "good" at the halfway point and "awful" after finishing the series. But a 3-D graph would be too complex so I'd have to find a simpler representation for this data. |
||||
|
||||
|
dormcat
Encyclopedia Editor
 Posts: 9902 Location: New Taipei City, Taiwan, ROC |
|
|||
3-D graphs are indeed too complex for both programmers and readers. How about having two separate bar graphs, one for "seen some" and one for "seen all?" A "will not finish" bar graph is optional, but that one is expected to be very lowly rated and does not really contributes to the overall rating. However, if a title has a significant portion of "will not finish" then it's likely to be crappy. There are easier things to do, such as additional languages... |
||||
|
||||
Mylene
 Posts: 2792 Location: Indiana |
|
|||
However, "Will not finish" does not necessarily indicate utter dislike of a series by any stretch of the imagination. Some NA Kodocha fans might shift the series into "Will not finish" simply because they will (may?) not have the opportunity to ever finish the series. I put several series such as Voltron into my "Will not finish," not because I hated them too much to continue, but because while I liked them in the past and gave them an average rating, I don't see myself going out of the way to watch every single episode. |
||||
|
||||
 |
All times are GMT - 5 Hours |
|
|
|
Powered by phpBB © 2001, 2005 phpBB Group