Loading ...
Sorry, an error occurred while loading the content.

Re: Baysian distributed reputation clusters? Not likely...

Expand Messages
  • gaijin@yha.att.ne.jp
    ... what s ... This inferred trusts (recommendations) is what NeuroGrid is trying to achieve. As search software it relies on building up trust about which
    Message 1 of 28 , Mar 31, 2001
    • 0 Attachment
      --- In decentralization@y..., "Bill Kearney" <wkearney99@h...> wrote:
      >
      >
      > Not trust without some kind of approval but as a guide.
      >
      > Quite a sticky mess to attempt to code. But it seems like a really
      > good place for pseudo-decentralized software.
      >
      > Maybe if we start with inferred trusts (recommendations) based on
      > loosely coupled groups we'll be able to get a good picture of
      what's
      > really *needed* not just what's possible.
      >
      > Thoughts?

      This "inferred trusts (recommendations)" is what NeuroGrid is
      trying to achieve. As search software it relies on building up trust
      about which other search engines can provide good results on which
      topics. When those other search engines make suggestions about
      things (actual data or even other search engines) they can be
      weighted to take into account the degree of trust that your personal
      NeuroGrid node or profile has learnt about them.

      I am now having discussions about using NeuroGrid as a basis for a
      FAQ system. NeuroGrid currently works with URLs and handles
      bookmarks but we are looking at a version that will associate
      questions with various possible answers (themselves associated
      with keywords) and allow a user to search through the sets of
      possible answers. This would allow NeuroGrid to work out on a
      user by user basis how well each answer matched each
      question and sort accordingly for future searches.

      NeuroGrid maintains data on all users actions separately so as well
      as getting an indication of which answers were most useful to the
      group as a whole you can work out which answers were most useful
      to which sub-groups and that sub-group can be selected by you
      (i.e. your friends) or inferred based on those who have preferred
      similar answers to you and have supplied you with useful results
      (through their NeuroGrid node) on previous occasions.

      CHEERS> SAM

      http://www.neurogrid.com
    • Clay Shirky
      ... Does this matter? In a high-volume situation like /., I d argue that the rough justice of the present system works well, since anything that gets a 4 or 5
      Message 2 of 28 , Apr 1, 2001
      • 0 Attachment
        > The unfortunate thing about slashdot is the failure of moderation. I
        > can't quite articulate this well but basically I'm finding a lot of
        > useful things aren't being moderated up.

        Does this matter? In a high-volume situation like /., I'd argue that
        the rough justice of the present system works well, since anything
        that gets a 4 or 5 tends to be at least informative, and most of the
        crap doesn't make it past 3.

        Any mod situation balances misses that should be hits with hits that
        should be misses, and as volume increases the mass of users tend to
        become more worried about being shown bad stuff than missing good
        stuff. For me, /. is easily at the point where I am more concerned
        that (modded up == good) than that (good == modded up).

        Similarly, one of the reasons Google won over Yahoo's implementation
        of Inktomi was that Yahoo was default boolean OR with substring
        matches, while Google was boolean AND with full word matches.

        (Though of course on /., some of this effect comes from the low TTL
        and write-only nature of a topic over a couple of hours old, which
        means that there simply arent't enough moderators in hours 2-N to
        raise good posts, so there is a kind of human 'signal decay' going on
        as well.)

        -clay
      • Lucas Gonze
        ... The problem I have with Advogato is that (1) it is about reputation of the writer instead of reputation of the post and (2) it s a centralized metric. (see
        Message 3 of 28 , Apr 2, 2001
        • 0 Attachment
          > Any mod situation balances misses that should be hits with hits that
          > should be misses, and as volume increases the mass of users tend to
          > become more worried about being shown bad stuff than missing good
          > stuff. For me, /. is easily at the point where I am more concerned
          > that (modded up == good) than that (good == modded up).

          The problem I have with Advogato is that (1) it is about reputation of the
          writer instead of reputation of the post and (2) it's a centralized metric.
          (see interesting thread at http://www.advogato.org/article/261.html).

          The most successful moderation system I can think of is webloggers posting links
          to stuff they like. The cool thing about it is that choosing good links
          enhances the blogger's reputation, which is a better incentive to moderate well
          than on /. What do you get for moderating well on slashdot?

          - Lucas
        • Bill Kearney
          ... on ... Indeed, you ve hit on very valid point. One I think is usually the basis for any calls for categorization or rating. The issue of an increasing
          Message 4 of 28 , Apr 2, 2001
          • 0 Attachment
            > Does this matter? In a high-volume situation like /., I'd argue that
            > the rough justice of the present system works well, since anything
            > that gets a 4 or 5 tends to be at least informative, and most of the
            > crap doesn't make it past 3.
            >
            > Any mod situation balances misses that should be hits with hits that
            > should be misses, and as volume increases the mass of users tend to
            > become more worried about being shown bad stuff than missing good
            > stuff. For me, /. is easily at the point where I am more concerned
            > that (modded up == good) than that (good == modded up).
            >
            > Similarly, one of the reasons Google won over Yahoo's implementation
            > of Inktomi was that Yahoo was default boolean OR with substring
            > matches, while Google was boolean AND with full word matches.
            >
            > (Though of course on /., some of this effect comes from the low TTL
            > and write-only nature of a topic over a couple of hours old, which
            > means that there simply arent't enough moderators in hours 2-N to
            > raise good posts, so there is a kind of human 'signal decay' going
            on
            > as well.)

            Indeed, you've hit on very valid point. One I think is usually the
            basis for any calls for categorization or rating. The issue of an
            increasing volume of users worried about missing good stuff. IMHO,
            most folks don't really want categorization. They're more interested
            in following "important" trends/topic. The determination
            of "importance" is the tricky part. Categorization is but one way to
            accomplish it. I'm assuming categorization gets considered because
            of it's "neutral" nature. Categories being theoretically separate
            from "opinion".

            My question is how do these opinions and categories get applied to
            news items in a way that fits my interests? Being able to "feed
            back" personal opinions (ratings?) of news items in a manner that is
            of interest to others seems to be missing. Sure, once
            someone "discovers" a source of news it may become part of their
            regular reading. But that discovery process seems unreasonably
            chaotic and certainly inefficient. One result of this is
            the "slashdot effect" when some nugget of info rises to the top and
            the teeming hordes descend upon it. This seems similarly problematic
            for distributed searching. In that you either cast too small a net
            and get no valid results or you crucify the network by casting too
            wide a search.

            If the data desired can be found based on common interests then how
            about a means to feed back into the system?

            -Bill Kearney
          • Todd Boyle
            ... What if you had an application, where you could select a bunch of text and rightclick, and score it from 1 - 10 from strongly negative, to strongly
            Message 5 of 28 , Apr 2, 2001
            • 0 Attachment
              > The problem I have with Advogato is that (1) it is about reputation
              > of the writer instead of reputation of the post and (2) it's a
              > centralized metric...( http://www.advogato.org/article/261.html).

              > The most successful moderation system I can think of is
              > webloggers posting links to stuff they like. The cool thing
              > about it is that choosing good links enhances the blogger's
              > reputation, which is a better incentive to moderate well
              > than on /. What do you get for moderating well on slashdot?
              >
              > - Lucas

              What if you had an application, where you could select a bunch
              of text and rightclick, and score it from 1 - 10 from strongly
              negative, to strongly positive. What if the app. would snarf up
              the metadata about the art. and send a message to the host.

              The negatively rated stuff text be printed in progressively
              lighter font until it faded out completely, like "Back to the Future"...

              But it would have to work like Mojonation, you can't vote unless
              you have some reputation mojo, to begin with. I suppose that
              reputation could come from getting lots of positive rightclicks.

              And like every other currency this "moxie" currency would be
              convertible to other currencies if you had a GL table as the
              data structure. thus a writer could make a living, I suppose,
              by selling moxie.

              This would help the big corporations corrupt the content like
              they do in the newspapers and on TV...

              Hmmm. Even if the moxie weren't convertible they could just
              tell the author what content to corrupt with his huge reputation.
              That would be more like today's media.

              This all boils down to a sort of populism in the long run...
              I already know what America thinks about almost any given
              issue and I'm not real impressed with either the average
              intelligence, or values, in America. Accordingly I'm not real
              interested in any truly democratic reputation system, frankly.

              I wonder what it would take to design a system that had true
              "intelligence". Is there such a thing?

              Maybe I want to Rightclick and select stuff I agree with, and
              the discussion platform would steer me towards

              A. more content like that, and
              B. more people like that.

              I suppose this is like amplification of prions but what the hell.
              It's no better amplifier than church on sunday,

              Todd
            • Gen Kanai
              ... I like MetaFilter.com for that reason. Small (under 5K) community, usually good content, usually good commentary, not only technology. But there isn t
              Message 6 of 28 , Apr 2, 2001
              • 0 Attachment
                At 12:18 PM 4/2/01 -0400, Lucas Gonze wrote:

                >The most successful moderation system I can think of is webloggers posting
                >links
                >to stuff they like. The cool thing about it is that choosing good links
                >enhances the blogger's reputation, which is a better incentive to moderate
                >well
                >than on /. What do you get for moderating well on slashdot?

                I like MetaFilter.com for that reason. Small (under 5K) community, usually
                good content, usually good commentary, not only technology.

                But there isn't anything on MeFi to rate posts or posters (a la Epinions)
                but it does get tedious to have to rate each thing you read or each post or
                comment. Something more seamless would be desirable- if there was a way to
                judge the time someone spent on a page or that X number of users commented
                = higher quality or higher interest.

                Gen
              • Bill Kearney
                ... post or ... a way to ... commented ... This hits the nail square on the head. The tedium of things greatly reduces the likelihood of obtaining valid data.
                Message 7 of 28 , Apr 2, 2001
                • 0 Attachment
                  > but it does get tedious to have to rate each thing you read or each
                  post or
                  > comment. Something more seamless would be desirable- if there was
                  a way to
                  > judge the time someone spent on a page or that X number of users
                  commented
                  > = higher quality or higher interest.

                  This hits the nail square on the head. The tedium of things greatly
                  reduces the likelihood of obtaining valid data.

                  I'm not sure that 'time on screen' or other session oriented metrics
                  are valuable. Many times a screen not read immediately is later
                  shown from cache. Or a page left onscreen whilst performing other
                  tasks does not represent any greater interest. I'd resent software
                  that acted that stupid.

                  If you want user opinion, make it *easy* for them to give it. (Now,
                  some folks may well NOT want my personal opinion but that's another
                  matter.) There seems to be a reasonable number of possible voting
                  categories or grades; pass/fail, 1 to 10, strongly/mildly
                  agree/disagree, etc. It would be interesting to see products start
                  to implement this sort of feature. Give the author/editor/publisher
                  a chance to request feedback. A simple start would be nothing more
                  than a URL back to a voting page that did NOT harass the user with an
                  avalanche of marketing tripe.

                  -Bill Kearney
                • Court Demas
                  I m afraid that any manual voting system would be too much of a hinderance for most users. It simply wouldn t be kept current. What s the most popular
                  Message 8 of 28 , Apr 3, 2001
                  • 0 Attachment
                    I'm afraid that any manual voting system would be too much of a hinderance for most users. It simply wouldn't be kept current.

                    What's the most popular reputation system today? Maybe google? It's quite rough, but linking to a page certainly carries a lot meaning. Unfortunately you're missing the majority of users who never create new pages (I wonder what that percentage is...)

                    I'd like to see a Groove-style small scale collaborative filtering mechanism. Something with me and a dozen of my friends (ecircles-like) where we share information on which Web pages we've seen, which we got to every day, which emails we read:

                    - add a local http proxy filter which tracks where I surf (I *want* this)
                    - share links with my group. links which my friends have spent lots of time are tagged <blink> (haha)
                    - some process watches my mailbox, and sees which messages I put into the "spam" folder. others in my group get a list of "urls in messages I thought were spam".

                    I've put effort into particular voting systems at times. For example, the excellent NetFlix (http://www.netflix.com/) services has a wonderful movie rating system. Unfortunately they seem to have lost my ratings at least once, and it took a lot of selfish motivation (wanting a movie recommendation) to sit there and tell them about 50 movies I'd seen.

                    So how else can we passively track what people are interested in, in a way that can be shared with others?

                    court

                    --- In decentralization@y..., "Bill Kearney" <wkearney99@h...> wrote:
                    > > but it does get tedious to have to rate each thing you read or each
                    > post or
                    > > comment. Something more seamless would be desirable- if there was
                    > a way to
                    > > judge the time someone spent on a page or that X number of users
                    > commented
                    > > = higher quality or higher interest.
                    >
                    > This hits the nail square on the head. The tedium of things greatly
                    > reduces the likelihood of obtaining valid data.
                    >
                    > I'm not sure that 'time on screen' or other session oriented metrics
                    > are valuable. Many times a screen not read immediately is later
                    > shown from cache. Or a page left onscreen whilst performing other
                    > tasks does not represent any greater interest. I'd resent software
                    > that acted that stupid.
                    >
                    > If you want user opinion, make it *easy* for them to give it. (Now,
                    > some folks may well NOT want my personal opinion but that's another
                    > matter.) There seems to be a reasonable number of possible voting
                    > categories or grades; pass/fail, 1 to 10, strongly/mildly
                    > agree/disagree, etc. It would be interesting to see products start
                    > to implement this sort of feature. Give the author/editor/publisher
                    > a chance to request feedback. A simple start would be nothing more
                    > than a URL back to a voting page that did NOT harass the user with an
                    > avalanche of marketing tripe.
                    >
                    > -Bill Kearney
                  • Julian Bond
                    In article , Court Demas writes ... Somebody out there has a system that builds a virtual proxy out of a workgoup s
                    Message 9 of 28 , Apr 3, 2001
                    • 0 Attachment
                      In article <9ac303+ba52@...>, Court Demas <court@...> writes
                      >- add a local http proxy filter which tracks where I surf (I *want* this)

                      Somebody out there has a system that builds a virtual proxy out of a
                      workgoup's browser's caches. Taking this a step further, there's a lot
                      of information to be mined from a workgoup's daily internet activity.
                      The webpages they visit, the mailing lists and newsgroups they post to.
                      The music they listen to.

                      I like the idea of building recommendation systems purely passively from
                      normal activity. But it's really hard to make them intelligent enough.
                      To take just one case, the company website is likely to get the most
                      hits as everyone has it as their home page. This would make it
                      disproportionately "interesting".

                      If this is done outside the workgoup there's lots of potential for
                      privacy abuse as well.

                      --
                      Julian Bond eMail: julian@...
                      HomeURL: http://www.shockwav.demon.co.uk/
                      WorkURL: http://www.netmarketseurope.com/
                      WebLog: http://roguemoon.manilasites.com/
                      M: +44 (0)77 5907 2173 T: +44 (0)20 7420 4363
                      ICQ:33679668 tag:So many words, so little time
                    • Jeff Barr
                      Bill says, ... Indeed! And this is why www.HotOrNot.com is a nice example of how to do evaluation without adding complexity to the UI. By combining the
                      Message 10 of 28 , Apr 3, 2001
                      • 0 Attachment
                        Bill says,

                        > This hits the nail square on the head. The tedium of things greatly
                        > reduces the likelihood of obtaining valid data.

                        Indeed! And this is why www.HotOrNot.com is a nice example of how
                        to do evaluation without adding complexity to the UI.

                        By combining the controls for "next picture" and "evaluate picture",
                        you can't help but give feedback. You don't get to be a passive
                        observer. You have to pay (rank) to play.

                        Jeff;

                        (And the pictures aren't bad either :-).

                        -----Original Message-----
                        From: Bill Kearney [mailto:wkearney99@...]
                        Sent: Monday, April 02, 2001 5:43 PM
                        To: decentralization@yahoogroups.com
                        Subject: [decentralization] Re: moderation


                        > but it does get tedious to have to rate each thing you read or each
                        post or
                        > comment. Something more seamless would be desirable- if there was
                        a way to
                        > judge the time someone spent on a page or that X number of users
                        commented
                        > = higher quality or higher interest.

                        This hits the nail square on the head. The tedium of things greatly
                        reduces the likelihood of obtaining valid data.

                        I'm not sure that 'time on screen' or other session oriented metrics
                        are valuable. Many times a screen not read immediately is later
                        shown from cache. Or a page left onscreen whilst performing other
                        tasks does not represent any greater interest. I'd resent software
                        that acted that stupid.

                        If you want user opinion, make it *easy* for them to give it. (Now,
                        some folks may well NOT want my personal opinion but that's another
                        matter.) There seems to be a reasonable number of possible voting
                        categories or grades; pass/fail, 1 to 10, strongly/mildly
                        agree/disagree, etc. It would be interesting to see products start
                        to implement this sort of feature. Give the author/editor/publisher
                        a chance to request feedback. A simple start would be nothing more
                        than a URL back to a voting page that did NOT harass the user with an
                        avalanche of marketing tripe.

                        -Bill Kearney


                        To unsubscribe from this group, send an email to:
                        decentralization-unsubscribe@egroups.com



                        Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
                      • Clay Shirky
                        ... Even better (exempting the rank sexism) is www.pickthehottie.com, which uses (Lucas, are you listening) bubble sort rather than absurd numeric scales, and
                        Message 11 of 28 , Apr 3, 2001
                        • 0 Attachment
                          > Indeed! And this is why www.HotOrNot.com is a nice example of how
                          > to do evaluation without adding complexity to the UI.
                          >
                          > By combining the controls for "next picture" and "evaluate picture",
                          > you can't help but give feedback. You don't get to be a passive
                          > observer. You have to pay (rank) to play.

                          Even better (exempting the rank sexism) is www.pickthehottie.com,
                          which uses (Lucas, are you listening) bubble sort rather than absurd
                          numeric scales, and thus makes the picking, not even the ranking, as
                          the cue to go to the next page.

                          I have an as-yet unformulated objection to all 'rank by number'
                          systems, having worked oin firefly these many years ago. I think in
                          particular that numeric ideas of transitive trust are wrong -- I don't
                          trust my mother "99%" or anything of the sort. Trust is both more
                          binary (I trust you or I don't, in a particular context) and grayer (I
                          vary my trust based on my assessment of the other demands on your time
                          and effort) than mere numerica scales suggest.

                          It would be great if human trust devolved to transitive integers, but
                          it isn't that simple. I think the "Explicit numerical ranking" crew is
                          over-computerizing the problem. When all you have is a number,
                          everything loooks like a scale.

                          -clay
                        • Justin Chapweske
                          I think to sum up your opinion Clay: Distributed ranking without distributed context is meaningless. ... -- Justin Chapweske, Lead Swarmcast Developer,
                          Message 12 of 28 , Apr 3, 2001
                          • 0 Attachment
                            I think to sum up your opinion Clay:

                            Distributed ranking without distributed context is meaningless.

                            >
                            > Even better (exempting the rank sexism) is www.pickthehottie.com,
                            > which uses (Lucas, are you listening) bubble sort rather than absurd
                            > numeric scales, and thus makes the picking, not even the ranking, as
                            > the cue to go to the next page.
                            >

                            --
                            Justin Chapweske, Lead Swarmcast Developer, OpenCola Inc.
                            http://www.sourceforge.net/projects/swarmcast/
                          • Lucas Gonze
                            ... excellent. Really excellent. Picking one or the other is fun, whereas picking a number between 1 and 10 is a little stressful. Plus you get to click on
                            Message 13 of 28 , Apr 3, 2001
                            • 0 Attachment
                              > Even better (exempting the rank sexism) is www.pickthehottie.com,
                              > which uses (Lucas, are you listening) bubble sort rather than absurd
                              > numeric scales, and thus makes the picking, not even the ranking, as
                              > the cue to go to the next page.

                              excellent. Really excellent. Picking one or the other is fun, whereas picking
                              a number between 1 and 10 is a little stressful. Plus you get to click on the
                              hotties.

                              > It would be great if human trust devolved to transitive integers, but
                              > it isn't that simple. I think the "Explicit numerical ranking" crew is
                              > over-computerizing the problem. When all you have is a number,
                              > everything loooks like a scale.

                              I would say that the actual number is more like a bid than a characteristic.
                              The question isn't whether your mom's trustability ranking is X, it's whether
                              you should take a chance that something will happen. That's as fuzzy as any
                              pricing mechanism. "I'll bid $X on this project, given the likely time to do
                              the work and likelyhood that the client will pay up." It all comes down to the
                              buyer and seller finding numbers that they can live with.

                              The reason this is numbers generated by computers and not just human gut
                              instinct is because computers are already engaged in economic transactions. The
                              point isn't to invent an economy from whole cloth, it's to allow an existing
                              economy to do business better.

                              - Lucas
                            • Lucas Gonze
                              ... explain?
                              Message 14 of 28 , Apr 4, 2001
                              • 0 Attachment
                                > Distributed ranking without distributed context is meaningless.

                                explain?
                              • Justin Chapweske
                                ... amihotornot provides no context for the users to keep their ratings consistant, basically I may have much lower standards than the next person as to what I
                                Message 15 of 28 , Apr 4, 2001
                                • 0 Attachment
                                  On Wed, Apr 04, 2001 at 10:55:56AM -0400, Lucas Gonze wrote:
                                  > > Distributed ranking without distributed context is meaningless.
                                  >
                                  > explain?
                                  >

                                  amihotornot provides no context for the users to keep their ratings
                                  consistant, basically I may have much lower standards than the next person
                                  as to what I consider hot.

                                  pickthehottie on the other hand provides a context within to make the
                                  decision because it is a relative choice (context) rather than an absolute
                                  numerical choice (no context).


                                  --
                                  Justin Chapweske, Lead Swarmcast Developer, OpenCola Inc.
                                  http://www.sourceforge.net/projects/swarmcast/
                                • Justin Chapweske
                                  One does not need the entire global context/state in order to make a decision. The most optimal solutions may require the entire state, but there will be
                                  Message 16 of 28 , Apr 4, 2001
                                  • 0 Attachment
                                    One does not need the entire global context/state in order to make a
                                    decision. The most optimal solutions may require the entire state, but
                                    there will be pretty good solutions that only need partial state.

                                    I think pickthehottie provides a pretty good example of this in that the
                                    users are not required to each download the entire archive of images and
                                    then set about the laborious task of ranking each one. Instead the users
                                    make a bunch of simple decisions based on a minimal amount of context.

                                    > On Wed, Apr 04, 2001 at 11:49:25AM -0700, Saman Faraz wrote:
                                    > Hi all, elbowing in- that's exactly the issue. The problem is that building
                                    > up a global context in a distributed system can be very expensive. The
                                    > aggregation of the info to construct can be viewed as a distributed merge
                                    > problem with high inter-process comm. delays.
                                    >
                                    > Saman
                                    >
                                    > on 4/3/01 11:10 AM, Justin Chapweske at orasis@... wrote:
                                    > > I think to sum up your opinion Clay:
                                    > >
                                    > > Distributed ranking without distributed context is meaningless.
                                    > >
                                    > >>
                                    >
                                    >
                                    > To unsubscribe from this group, send an email to:
                                    > decentralization-unsubscribe@egroups.com
                                    >
                                    >
                                    >
                                    > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
                                    >

                                    --
                                    Justin Chapweske, Lead Swarmcast Developer, OpenCola Inc.
                                    http://www.sourceforge.net/projects/swarmcast/
                                  • Saman Faraz
                                    Hi all, elbowing in- that s exactly the issue. The problem is that building up a global context in a distributed system can be very expensive. The
                                    Message 17 of 28 , Apr 4, 2001
                                    • 0 Attachment
                                      Hi all, elbowing in- that's exactly the issue. The problem is that building
                                      up a global context in a distributed system can be very expensive. The
                                      aggregation of the info to construct can be viewed as a distributed merge
                                      problem with high inter-process comm. delays.

                                      Saman

                                      on 4/3/01 11:10 AM, Justin Chapweske at orasis@... wrote:
                                      > I think to sum up your opinion Clay:
                                      >
                                      > Distributed ranking without distributed context is meaningless.
                                      >
                                      >>
                                    • Clay Shirky
                                      ... I was bitching at Lucas about this just the other day. Those people who are always wringing their hands about finding global maxima always forget to add
                                      Message 18 of 28 , Apr 4, 2001
                                      • 0 Attachment
                                        > One does not need the entire global context/state in order to make a
                                        > decision. The most optimal solutions may require the entire state, but
                                        > there will be pretty good solutions that only need partial state.

                                        I was bitching at Lucas about this just the other day. Those people
                                        who are always wringing their hands about finding global maxima always
                                        forget to add the "time to find solution" axis to their maps. Anyone
                                        who has ever gone on a date understands how local maxima *are* global
                                        maxima considered as a factor of time.

                                        > I think pickthehottie provides a pretty good example of this in that the
                                        > users are not required to each download the entire archive of images and
                                        > then set about the laborious task of ranking each one. Instead the users
                                        > make a bunch of simple decisions based on a minimal amount of context.

                                        Its not even minimal context thats the issue, its _local_ context that
                                        makes PTH superior to HotorNot. It recreates the kind of real world
                                        choices for which our brains are already optimized -- coffee or tea,
                                        the chicken or the pasta, sleep or more Quake? rather than 'Rate Coke,
                                        coffee, Thera-flu, and martinis each on a scale of 1-7.'

                                        -clay
                                      • James Hong
                                        We considered using a similar voting model when we started HOTorNOT, as it is superior in many ways, such as the fact that it enables you to get two votes out
                                        Message 19 of 28 , Apr 4, 2001
                                        • 0 Attachment
                                          We considered using a similar voting model when we started HOTorNOT, as it
                                          is superior in many ways, such as the fact that it enables you to get two
                                          votes out of one.

                                          However, we feel there are some bad things about the model too. One is that
                                          we speculate it may be *more* mentally taxing for a user to have to decide
                                          between two choices. A lot of people hate making decisions (e.g. the people
                                          in front of you at McDonalds who sit there all day saying "i'll have a
                                          hamburger.. no, no, umm, make that a chicken sandwich.. um... shoot.. no,
                                          make it a quarter pounder) What if the two pictures are "about the same",
                                          which do they pick? HOTorNOT was designed to be somewhat "mindless" and
                                          simple, and throwing in TWO pictures makes is less so. We felt because the
                                          1-10 rating model is so commonplace in our society, it would require less
                                          thought on the user's part.

                                          More importantly, the Pickthehottie model also steers the website into being
                                          a competition, which is something we do not want to do. Our goal was to make
                                          our site "fun, clean, and real" for both the submitter and submitee, and I
                                          can guarantee you that making it a competition decreases a person's
                                          propensity to submit.

                                          Anyhow, I agree that the model is more efficient and attractive in theory,
                                          but for this application I just wanted to point out that it has its
                                          downsides as well.

                                          cheers,
                                          james

                                          ---
                                          XMethods web service listings - http://www.xmethods.net

                                          ----- Original Message -----
                                          From: "Clay Shirky" <clay@...>
                                          To: <decentralization@yahoogroups.com>
                                          Sent: Wednesday, April 04, 2001 1:42 PM
                                          Subject: Re: [decentralization] Re: moderation


                                          > > One does not need the entire global context/state in order to make a
                                          > > decision. The most optimal solutions may require the entire state, but
                                          > > there will be pretty good solutions that only need partial state.
                                          >
                                          > I was bitching at Lucas about this just the other day. Those people
                                          > who are always wringing their hands about finding global maxima always
                                          > forget to add the "time to find solution" axis to their maps. Anyone
                                          > who has ever gone on a date understands how local maxima *are* global
                                          > maxima considered as a factor of time.
                                          >
                                          > > I think pickthehottie provides a pretty good example of this in that the
                                          > > users are not required to each download the entire archive of images and
                                          > > then set about the laborious task of ranking each one. Instead the
                                          users
                                          > > make a bunch of simple decisions based on a minimal amount of context.
                                          >
                                          > Its not even minimal context thats the issue, its _local_ context that
                                          > makes PTH superior to HotorNot. It recreates the kind of real world
                                          > choices for which our brains are already optimized -- coffee or tea,
                                          > the chicken or the pasta, sleep or more Quake? rather than 'Rate Coke,
                                          > coffee, Thera-flu, and martinis each on a scale of 1-7.'
                                          >
                                          > -clay
                                          >
                                          >
                                          >
                                          >
                                          >
                                          > To unsubscribe from this group, send an email to:
                                          > decentralization-unsubscribe@egroups.com
                                          >
                                          >
                                          >
                                          > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
                                          >
                                        • Clay Shirky
                                          ... I totally buy making HotorNot less competitive, but I don t get this point. Are you saying people would find it easier to decide at McD s if they ranked
                                          Message 20 of 28 , Apr 4, 2001
                                          • 0 Attachment
                                            > However, we feel there are some bad things about the model too. One is that
                                            > we speculate it may be *more* mentally taxing for a user to have to decide
                                            > between two choices. A lot of people hate making decisions (e.g. the people
                                            > in front of you at McDonalds who sit there all day saying "i'll have a
                                            > hamburger.. no, no, umm, make that a chicken sandwich.. um... shoot.. no,
                                            > make it a quarter pounder).

                                            I totally buy making HotorNot less competitive, but I don't get this
                                            point. Are you saying people would find it easier to decide at McD's
                                            if they ranked every meal on a scale of 1-10 and then took the food
                                            with the highest numerical ranking?

                                            > What if the two pictures are "about the same", which do they pick?

                                            The one they think is better looking, of course. People make this kind
                                            of distinction all day every day -- two apples in the store, two seats
                                            on the bus, two lines at the bank. I am arguing, in fact, that we are
                                            fantastically good at choosing between 2 or more very similar things
                                            that are present than ranking a single instance of N things on an
                                            uncalibrated scale that must accomodate all N, known and unknown.

                                            I'm not arguing that PTH is the more successful _site_, mind you; HoN
                                            obviously wins hands down in that case, not least because of the
                                            perverse desire to see how one aligns with global opinion. In terms of
                                            a ranking application, though, I can't see how bubble-sort seems
                                            *harder* to you than ranking on an explicit but undefined scale.

                                            -clay
                                          • James Hong
                                            ... that ... decide ... people ... no, ... Actually, all I am trying to say (albeit badly) is that I think people find it easier to rank PEOPLE from 1-10, and
                                            Message 21 of 28 , Apr 4, 2001
                                            • 0 Attachment
                                              > > However, we feel there are some bad things about the model too. One is
                                              that
                                              > > we speculate it may be *more* mentally taxing for a user to have to
                                              decide
                                              > > between two choices. A lot of people hate making decisions (e.g. the
                                              people
                                              > > in front of you at McDonalds who sit there all day saying "i'll have a
                                              > > hamburger.. no, no, umm, make that a chicken sandwich.. um... shoot..
                                              no,
                                              > > make it a quarter pounder).
                                              >
                                              > I totally buy making HotorNot less competitive, but I don't get this
                                              > point. Are you saying people would find it easier to decide at McD's
                                              > if they ranked every meal on a scale of 1-10 and then took the food
                                              > with the highest numerical ranking?

                                              Actually, all I am trying to say (albeit badly) is that I think people find
                                              it easier to rank PEOPLE from 1-10, and only because the concept of rating a
                                              person from 1-10 is so established, not because it is intrinsically easier
                                              or better. The "learning curve" our users face to understand what our
                                              website is about is low because people already understand the 1-10 concept
                                              for people. We don't have to teach them anything.

                                              It is not as common (at least for people like me), however, to be able to
                                              pick and choose people, so people are not *as* conditioned to compare
                                              people. For instance, if you see a beautiful woman, you don't immediately
                                              think "wow, she's more beautiful than the girl standing next to her", you
                                              think "wow, she's gorgeous! a perfect 10". When you are dealing with the
                                              rating of objects in which people generally DO have options and commonly
                                              make choices, the opposite is true and the bubble-sort algorythm would be
                                              better. (e.g. I would rather have a chicken sandwich than a big mac).

                                              One other thing to point out is that the bubble sort is not better when
                                              there is information in the score beyond the average...In the case of looks,
                                              different people have very different opinions. A person's votes on HON are
                                              ALWAYS a bell curve. If we were to use the bubble sort, the person who got
                                              an average score of 6 would not realize that there were people out there who
                                              gave him 8's, 9's, and maybe 10's. That data is of tremendous value to our
                                              users, and is one advantage of the 1-10 system.


                                              >
                                              > > What if the two pictures are "about the same", which do they pick?
                                              >
                                              > The one they think is better looking, of course. People make this kind
                                              > of distinction all day every day -- two apples in the store, two seats
                                              > on the bus, two lines at the bank. I am arguing, in fact, that we are
                                              > fantastically good at choosing between 2 or more very similar things
                                              > that are present than ranking a single instance of N things on an
                                              > uncalibrated scale that must accomodate all N, known and unknown.
                                              >
                                              > I'm not arguing that PTH is the more successful _site_, mind you; HoN
                                              > obviously wins hands down in that case, not least because of the
                                              > perverse desire to see how one aligns with global opinion. In terms of
                                              > a ranking application, though, I can't see how bubble-sort seems
                                              > *harder* to you than ranking on an explicit but undefined scale.
                                              >

                                              As I think I mentioned (or implied), I think all things being equal, the
                                              bubble sort algorythm IS a more efficient and better methodology. I just
                                              wanted to point out, in our defense, that for our specific application there
                                              were other things we had to consider that led us to choosing the 1-10
                                              system.

                                              Anyhow, sorry, I didn't mean to take this conversation off track!

                                              -james
                                            • Saman Faraz
                                              ... true enough in many cases. Where it does become an issue though is when there isn t a reasonably uniform distribution of the sought after result/ranking
                                              Message 22 of 28 , Apr 4, 2001
                                              • 0 Attachment
                                                on 4/4/01 1:42 PM, Clay Shirky at clay@... wrote:

                                                >> One does not need the entire global context/state in order to make a
                                                >> decision. The most optimal solutions may require the entire state, but
                                                >> there will be pretty good solutions that only need partial state.
                                                >
                                                > I was bitching at Lucas about this just the other day. Those people
                                                > who are always wringing their hands about finding global maxima always
                                                > forget to add the "time to find solution" axis to their maps. Anyone
                                                > who has ever gone on a date understands how local maxima *are* global
                                                > maxima considered as a factor of time.

                                                true enough in many cases. Where it does become an issue though is when
                                                there isn't a reasonably uniform distribution of the sought after
                                                result/ranking criteria in the dataset.

                                                In my experience this seems to happen a fair bit, especially with
                                                needle-in-a-haystack problems, like with results ranking in partitioned
                                                repositories- if there are two databases and the query requests a ranked
                                                result of documents/records containing X and Y. If database 1 has documents
                                                with lots of X in them, and no Y, while database 2 has documents with lots
                                                of Xs and a few documents with X and Y.

                                                Database 2 can rank the X and Y documents at the top, but those can often
                                                get washed out during the merge by results from database 1 which have just
                                                a lot of X in them.

                                                Saman
                                              • Saman Faraz
                                                James, have you noticed any correlation between where the woman is in the line up and what her score ends up being? Saman
                                                Message 23 of 28 , Apr 4, 2001
                                                • 0 Attachment
                                                  James,
                                                  have you noticed any correlation between where the woman is in the line up
                                                  and what her score ends up being?
                                                  Saman
                                                  >
                                                  > More importantly, the Pickthehottie model also steers the website into being
                                                  > a competition, which is something we do not want to do. Our goal was to make
                                                  > our site "fun, clean, and real" for both the submitter and submitee, and I
                                                  > can guarantee you that making it a competition decreases a person's
                                                  > propensity to submit.
                                                  >
                                                • Lucas Gonze
                                                  FYI, folks, this group is completely moderated. That is because Yahoo s spam control is so bad that the only messages it ever flags as spam are ham. There
                                                  Message 24 of 28 , Jun 24, 2010
                                                  • 0 Attachment
                                                    FYI, folks, this group is completely moderated. That is because
                                                    Yahoo's spam control is so bad that the only messages it ever flags as
                                                    spam are ham. There are 800 fugly spams in the moderation queue right
                                                    now, with 799 of them spam, and two real messages in the spam queue,
                                                    both of which are ham.

                                                    So if you send a message and it doesn't go through, the reason is that
                                                    I haven't moderated it through yet. No need to resend, but if you get
                                                    frustrated waiting you might want to ping me directly.
                                                  Your message has been successfully submitted and would be delivered to recipients shortly.