Oct 072009
 

Randy Farmer (he of Habitat fame, and much more besides!) and Bryce Glass have been posting excerpts from their upcoming book Building Web Reputation Systems on a blog, and today’s has a great anecdote in it that hammers home all the math behind negative reputation systems.

“Hi! I see from your hub that you’re new to the area. Give me all your Simoleans or my friends and I will make it impossible to rent a house.”

“What are you talking about?”

“I’m a member of the Sims Mafia, and we will all mark you as untrustworthy, turning your hub solid red (with no more room for green), and no one will play with you. You have five minutes to comply. If you think I’m kidding, look at your hub-three of us have already marked you red. Don’t worry, we’ll turn it green when you pay…”

If you think this is a fun game, think again-a typical response to this shakedown was for the user to decide that the game wasn’t worth $10 a month. Playing dollhouse doesn’t usually involve gangsters.

– Building Web Reputation Systems: The Blog: The Dollhouse Mafia, or “Don’t Display Negative Karma”.

There’s whole rough drafts of chapters on the site — totally worth reading, pondering, absorbing, and using.

  23 Responses to “Don’t Display Negative Karma”

  1. That’s a great little article. They might have sold me on their book with it.

  2. I don’t know… the other day I was thinking about a reputation system entirely driven by displaying negative karma. Specifically, I was imagining what a forum community would look like if Facebook’s “like” feature or the typical “rate this post” feature was provided additional options, such as “rate this post as [narcissistic, dry, stupid, naive, pointless, etc.]” As a result, beneath the post, which users, how many users, and what they rated the post would be displayed. No anonymity.

    I think this would reduce the number of negative responses because of the availability of a negative feedback system a la “suggestion boxes.” Granted, there might also be a chilling effect on [stupid, etc.] posts, but hey, that’s not a bad thing. This might also encourage users to assume good faith of those who do post responses. Such a system might only work with certain communities, such as Something Awful. But, oh, how I wish Hulu and IMDb had negative feedback systems…

  3. Cool thanks I’ll check it out :)

  4. For the IMDb, what you need is not a list of opinions. What you need is a list of opinions of people whose opinions you share.

    In a first-order system to tackle this, you’d read reviews of movies you’d already seen, and if you disagreed with them you’d mark them bad. If you agreed with them, you’d mark them good. No-one else need see these ratings – they’d just be your opinion. For subsequent reviews, you would see reviews from reviewers you agreed with first, and ones from reviewers you disagreed with last.

    This is fine if there are lots of people who write many reviews. It doesn’t work so well if the person whose reviews you like only wrote 10 reviews, or there are 10 people you don’t like who only wrote 1 review. However, we can fix this by going to a second-order system.

    In a second-order system, your ratings of reviewers are compared with those of other users. If they coincide, then the views of these other users can be used as a proxy for your own views. If you and I generally like the same critics and dislike other critics, then the chances are that if you’ve rated a critic that I haven’t rated, I would agree with you. Therefore, your opinions help order the reviews that I see.

    Note that this doesn’t have to work on reviews – it could work for the movies themselves. If I rate a particular set of movies as ones I love, and some random person in Arkansas has rated most of those movies the same way I have, then if they’ve seen a movie they like and I haven’t, the chances are that I’ll like it too. I don’t need to know who this mystery reviewer is (in practice, there would be a set of them anyway); all I need to know is that they like the same kind of movies I do.

    This kind of shared rating system can work for MMOs, too. Indeed, it was in this context that I first came up with it (although I was almost certainly re-inventing the wheel – I asked Randy Farmer if he knew of such a system and he said he did). In the above Sims Mafia example, sure, they could rate you red, but your rating would only be seen as red by people whose opinions aligned with those of the Sims Mafia.

    Randy and Bryce have done an enormous amount of research for this book, and put a lot of thought into it. I’m really looking forward to seeing what their conclusions are.

    Richard

  5. The influence of the “mafia” griefers in The Sims Online got some minor press, but though they captured the attention of a gullible reporter or two, they only had a fraction of the impact that they imagined in their fevered griefer dreams. The reputation system was ignored by the bulk of the player base once it was apparent that it was prone to abuse (both on the positive and negative side of the scale). And obviously the scam outlined would fail miserably against anybody who’d been in the game long enough to locate the “Report Abuse” feature, as that function included integrated chat logging.

    That said, the point stands. The reputation system was rendered worse than useless, not by the griefers so much as the powergamers, who took it as the only meaningful “scoring” system in the game and played it as such.

    While Richard’s proposal is interesting, it assumes the baseline reviews are in good faith. It would still be too easy to bring in “clean” alts to do the wet-work, and run them through the delete-repeat cycle if they’re compromised.

    I’m (reluctantly) reaching the conclusion that the only reliable way to control bad actors in the virtual space is for companies to verify RL user identity for all accounts, and to share information on confirmed griefers in a central exchange. If you’re banned for scamming newbies in DDO, you might find your WoW and KOTOR characters have been deleted as well, or at least are flagged as potential troublemakers.

  6. It is subscription to authority, Richard. It is like rating the best schools (say Harvard, MIT, etc.) then giving preference to employment candidates with degrees from these institutions. Aka, choosing the chooser of choices (the real basis for power in society).

    It’s a rather old way to get it done. The difference now is dynamic assignments that potentially keep the rankings from settling into suboptimum minima as a result of positive feedback: ie, once a country star, always a country star over ‘what have you done for me lately’.

  7. Surely the hole in the model is the missing *right* to rank someone.

    The reason the Sims Mafia has power is because the rankings they bestow are equally weighted with the rankings of people ‘in-the-know’, even though they are ranking a complete stranger.

    Usually a ranking happens after some genuine interaction. In a web forum it would be reading a post, or on eBay after buying or selling something.

    Giving someone the ability to rank a stranger (good or bad) is surely the breakdown.

    Might be tough to come up with good metrics for who has an informed opinion about whom, though.

  8. The problem with that approach, Richard, is that you just end up promoting groupthink. The diversity of opinion is necessary, even if we don’t like what others have to say; we just need a way to express our dislike, so those feelings are effectively vented. Hiding that expression might have the same effect that suggestion boxes (that nobody checks) do, but I’d rather see notes such as “Jack and 6 other people thought this post was naive” below posts. Therapy through wisdom of the crowd.

  9. Note that this doesn’t have to work on reviews – it could work for the movies themselves. If I rate a particular set of movies as ones I love, and some random person in Arkansas has rated most of those movies the same way I have, then if they’ve seen a movie they like and I haven’t, the chances are that I’ll like it too. I don’t need to know who this mystery reviewer is (in practice, there would be a set of them anyway); all I need to know is that they like the same kind of movies I do.

    I’m surprised no one has pointed out that Netflix does this.

  10. BTW, there have been mathematical proofs done on negative rep systems showing that they always tend to spiral out of control, whereas positive rep systems start out that way and gradually get better.

  11. In terms of what I observe about human behavior, a considered compliment takes longer to make than an ill considered insult. You get fewer but they count for more. We take good behavior for granted. There is an age component to that as well.

    Yeah, Netflix gave out a prize for that one. A fun test of character is to find out how people react if they live in San Francisco or Los Angeles and discover they have the same tastes as someone in Delight, Arkansas.

  12. As a player of MMOs I have to say I’ve been hoping for some kind of negative karma system for some time.

    Say you play a random instanced adventure with 5 other people and one guy is a real jerk. If you and the others can flag him as someone unpleasant it would make the internet anonymity jerk syndrome much less prevalent.

    Obviously it could be abused but as long as it’s tied to some kind of shared activity that is intended to be cooperative I think it would do more good than harm.

    The guy who afk leeches in Alterac Valley really does deserve to have 39 players on his side give him negative feedback.

  13. Yukon Sam>While Richard’s proposal is interesting, it assumes the baseline reviews are in good faith. It would still be too easy to bring in “clean” alts to do the wet-work, and run them through the delete-repeat cycle if they’re compromised.

    Can you explain what you mean here? How can a clean alt be used to exploit this kind of system? It seems to me that the worst they can do is establish a set of honest responses, then do one bad one on a newbie.

    Richard

  14. Morgan Ramsay>The problem with that approach, Richard, is that you just end up promoting groupthink.

    But the group is of people who think the same way as you. It’s not that you just follow what the group thinks – you have to do some ratings yourself so you can find the right match. If your opinion is so diverse that no-one else rates the same way you do, OK, well in that case you’re not going to get any matches and you have to rely on your own notes instead.

    >we just need a way to express our dislike, so those feelings are effectively vented.

    What’s stopping you from doing this in the system I described? The more you state an opinion, the better the matches with other people’s votes are going to be.

    Michael Chui>I’m surprised no one has pointed out that Netflix does this

    I didn’t point it out because I’ve never heard of Netflix..!

    So does it work, then?

    Richard

  15. Raph: BTW, there have been mathematical proofs done on negative rep systems showing that they always tend to spiral out of control, whereas positive rep systems start out that way and gradually get better.

    I’d like to see the details of those proofs.

    Richard: So does it work, then?

    It seems to. I’ve been surprised and delighted by a lot of recommendations out of left field, and I haven’t heard anyone say anything bad about what Netflix suggests, which could mean anything. :)

  16. Can you explain what you mean here? How can a clean alt be used to exploit this kind of system? It seems to me that the worst they can do is establish a set of honest responses, then do one bad one on a newbie.

    Nice three, trash one, nice three, trash one… as long as the player and his gang keeps the ratio of normative responses high in relation to the character assassinations, they’ll be weighted the same as if they were ‘normal’ players, particularly in a system where most players don’t want to be bothered to rate others.

    I’m not familiar with the Netflix system, but I wonder how it would hold up to a situation such as that with Spore, where every review site was flooded with user comments and ratings trashing the game, primarily due to EA-Maxis’ ill-conisidered protection schema.

  17. For negative reputation to work, there has to be traceability and accountability. End of story.

  18. Correct. Otherwise it’s mau mau.

  19. Yukon Sam>Nice three, trash one, nice three, trash one… as long as the player and his gang keeps the ratio of normative responses high in relation to the character assassinations, they’ll be weighted the same as if they were ‘normal’ players, particularly in a system where most players don’t want to be bothered to rate others.

    This couldn’t happen with the system I outlined.

    Let’s say that you and I all rate A, B and C good. Your list will be in accordance with mine, so if you come across D first and trash them, yes, I’ll see a bad review. However, if I come across D first and rate them good, but then you trash D, our lists our now out of accord. Your list is not a good baseline for mine any more, because we have a disagreement. The only way it would be considered would be if my list was at odds with everyone else’s lists, so yours came up as a second-best match (and would be flagged as such).

    Richard

  20. If I’m reading this right, Bartle’s proposed system is similar to Netflix’s suggested titles or Amazon’s recommendations… the system builds a profile of each user based on the rankings they’ve been giving, and then it looks at how closely that profile matches the profile of other people, then provides greater weight to the rankings made by people similar to them. It’s not really gameable; your ratings only show up to people who have a similar rating pattern. If you try to deliberately mark a bunch of people down, you’ll end up with an atypical pattern and won’t weigh heavily on anyone else’s list.

    I’m not sure how well this would work in the MMOG space though; since you have an incomplete profile when you’re newer, newbies will end up denied the benefit of the ratings system. This is kinda bad, since predation usually ends up directed toward newbs, not elder players. They need to know, more so than anyone else, who is trust worthy and who is not. At it’s best, though, this system would be a way for older players to rate a newer player, or in really big games, it could be useful for rating players that are outside of any given player’s subcommunity, so there are definite benefits. You just need something else to balance out the fact that you lose the ability to provide guidance to newer players, but perhaps a strong mentoring system coupled with this might work?

  21. While the newer players may not have a profile to match against for gauging the reputation of their peers, they are at least somewhat shielded from their peers’ opinions of themselves being tainted. The only people who’s system-suggested opinion of that player will be likely to go negative will be other abusers. Perhaps a minimum number of votes and a minimum amount of time should need to transpire before their ‘ranking’ becomes publicly viewable. This could allow for a number of (legitimate) social interactions to take place, giving the new player a chance to be given an ‘in’ with the side of the playerbase that utilizes the system in an honest and straightforward manner.

    Isn’t it such fun outflanking and staying one step ahead of sociopaths?

  22. Kerri, absolutely, but you still need to deal with the new player integration angle, because sociopaths won’t game the rep system anymore, just abuse the fact that new players can’t tell that they’re sociopaths yet.

    There really needs to be some way of easing the players into communities so that they can be protected from that sort of predation.

    Though, come to think of it, you could probably significantly reduce the complexity of the system if it’s instituted on a community wide basis, and then you formalize community structures. Membership would have to allow for overlapping groups, but that’s not a huge deal. Instead of individual profiles, all you’d need to do is look at how the community members set ratings, which would only show up to members of the community. If there are people that try to abuse this system, they’d likely get pushed out of the community pretty fast, unless the community is a community of griefers.

  23. Actually, thinking on it further, the best place to start is the friend’s list. Tiering up for community involvement like guilds or formalized community structures expands your list out more, but for people not in those groups, the friends list is a sensible fall back.

Sorry, the comment form is closed at this time.