Wikipedia talk:Trust network/Archive

Just some quick thoughts - I've only skimmed the mailing list posts in question, so apologies if this has all been said already. I think the key point with webs of trust is that they are not purely quantitive, in the sense that they take into account who is doing the nominating - if somebody has a high trust score, their opinion is effective in giving trust scores to others; if they have a low trust score, their opinion will have little or no effect. This minimises the impact of "sock puppets", and is also I believe the key concept behind Google's famous PageRank system - links from a highly ranked site bestow more ranking on their target than links from a low-ranked site. Also, a key experiment in this area that people might want to look at is Advogato.

It all gets a bit recursive, but otherwise you can only go as far as personal black- and white-lists - like Slashdot's lists of "friends", "foes", "friends of friends", "foes of friends", etc. - IMSoP 16:51, 17 Feb 2004 (UTC)

As my mother once advised me, "You trust your mother, but you cut the cards." -- Jmabel 19:43, 17 Feb 2004 (UTC)

Hrmh! The page assumes "we all" know how to hack tables. I guess that is some sort of a "pons asinorum" to exclude newcomers, but while I have contributed for nearly 10 months now, I still do not grok tables or image-files. N.B. This is not really a comment about the concept itself. - Cimon Avaro on a pogostick 20:25, Feb 17, 2004 (UTC)


 * It would be good to have a format such that we experiment with the concept as widely as possible. If you can see an better layout, let's go for it. Pete/Pcb21 (talk) 09:00, 18 Feb 2004 (UTC)


 * P.S. Although I agree with you that tables can be intimidating, images are not that bad - maybe you should try harder over the next 10 months :-) Pete/Pcb21 (talk) 09:00, 18 Feb 2004 (UTC)

This may be a good idea to remove clutter from RC, but if its to be implemented it should be private. If, for example, a user doesn't trust another, and its announced publicly as such, it will inspire further aggravation and infighting between the users, which is probably counterproductive to editing in harmony. Dysprosia 23:38, 17 Feb 2004 (UTC)


 * If we just the trust network for the uncluttered RC, then making it private makes perfect sense. However if we use it for a global trust value, i.e. use the second table, then it has to be public otherwise a malevalent user could just set up a load of sock puppet accounts all of whom distrust a particular (good) user and break the system. (Although weightings might ameloriate this a bit, all the sock puppets might trust each other, giving them a heavy weighting - this was discussed/hinted at on the mailing list - basically private global trust scores are hard. Local ones, which will do for personalized RC, are not. Pete/Pcb21 (talk) 09:00, 18 Feb 2004 (UTC)


 * To some extent this can be addressed by having some "cost" to expressing trust for others. This is similar to google's algorithm for Page Rank. You start with a certain arbitrary "trust rating", which is increased slightly for each person who says he or she trusts you, and diminished slightly less for each person you say you trust or distrust. For this purpose, the algorithm has to be such that two people saying they trust each other slightly increases each of their trust levels. Yes, someone could set up sock puppets in a mutual trusting relationship (just like someone can use all sorts of illicit Search engine optimization techniques). One has to hope that (1) some people will indicate their distrust of the sock puppets and (2) wikipedia in general has ways of weeding out sock puppets. One possibility would be an algorithm requiring (by some measure) a certain number of non-trivial edits to be counted in this process at all; that would make relevant sock puppets more time-consuming to create. -- Jmabel 09:49, 18 Feb 2004 (UTC)

-

Neat page - thanks for setting it up. It just occurred to me that if we are able to largely get rid of the sock puppet problem this system could also be extended to articles in order to help set up Wikipedia 1.0. But as I stated before the web of trust should start as a more or less private thing that only affects what the users who choose to use it see. It can be extended later once/if we greatly minimize the potential for abuse. --mav 11:08, 18 Feb 2004 (UTC)


 * I agree that the local and private way of doing things is the way to go at least initially. As far as I can tell that system could be implemented relatively easily (especially if the "trust by proxy" bit is ignored as a very first step) - in the routine for RC it would say something like "for each entry if $USER->HAS_TRUSTED_LIST and $EDITOR is on $USER->TRUSTED_LIST then make entry $TRUSTED_COLOUR". There would be some cases having to deal with enhanced recent changes. As for the extended "global" scoring that Jimbo wants, I am sure we have the talent here amongst the wikipedians to figure an algorithm that would minimize the effects of sock puppets but it might take a while - lets go for simple, workable route first.


 * As for extending it to articles - I can see how this might work. I mark a version of an article "trusted" then as long as all subsequent edits are made by editors I trust (or who I trust by proxy), the article remains "trusted". If an unknown/untrusted user makes an edit it becomes untrusted by me until I mark it trusted again. Then Wikipedia 1.0 consists of those articles that, say, 95% of editors trust. I can generate MyWikipedia by exporting all articles that I trust (or trust by proxy).


 * All in all, this really means I must bite the bullet and go through the steep learning curve that is getting mediawiki set up on my PC so that I can help out with development. Pete/Pcb21 (talk) 11:58, 18 Feb 2004 (UTC)


 * Alternatively, we could use the connected network of user trusts starting with Jimbo as the centre - all pages trusted by voting users whom Jimbo (or whoever) trusts or trusts-by-proxy or trusts-by-proxy-by-proxy or ... are included (versions of said articles selected by modal selection).
 * Version 1.0 could then be (based on) the result of this network at a specific date (1st December? &c.), or at a certain number (1 million articles; &c.). But then we'd have the problem of people saying that certain trusted articles aren't worthy of being in WikiPediaPaperEdition(tm), and that we don't have enough (or any) coverage in area x....
 * Still, interesting...
 * James F. (talk) 16:45, 18 Feb 2004 (UTC)

A public listing of who you distrust seems little different to making a personal attack on those people. I'm worried it will only lead to bad feeling. Angela. 23:40, Feb 18, 2004 (UTC)
 * Yes such a public pronouncement would be highly dangerous in a live, implemented system. Indeed when setting up the example, I only dared distrust Pcb22, who is my sock puppet. Distrust would work fine in the private local idea. However this is another problem for the "global" score. It would become clear to some users that they are widely distrusted, but they wouldn't know who distrusts them, so can't seek to regain their trust! Yet another reason why this method is no good for a global score... a shame as that was the problem Jimbo was originally trying to solve.. but mav's slightly ambitious idea seems to be holding up to scrutiny, which is great. Pete/Pcb21 (talk) 09:40, 19 Feb 2004 (UTC)


 * Perhaps distrust would only be a private thing and not transfered via the web of trust to other users? --mav


 * What about open and closed (dis)trust? 5 options:
 * open trust
 * hidden trust
 * neutral
 * hidden distrust
 * open distrust
 * Where open distrust is only to be used for extreme cases... Guaka 7 July 2005 22:03 (UTC)


 * I think a problem with having any open trusts/distrusts is that it becomes too political. Let's say I think User:Angela is a good person to ally with (she's got a lot of people trusting her). If I want to make friends, I might think it wise to put her name up on my open trust list. This way 1) she'll see that I did so, and may be more likely to trust me, 2) other's will see that I did so, and infer that I'm on the side of the Good and True. Likewise, if I want to show how Good and True I am, I'll put the Evil User:Pcb22 on my list, whome no-one in the "in" crowd would be seen dead with. Finally, it will encourage arguments between those high-profile users and admins who are always having a go at each other &mdash; do I pick one? Do I pick the other? Who do I need to ally with? Panic!
 * I think the idea is great in its simplest possible form: a hidden list, editable in Preferences, of users whos edits in RC can be shown, shown lightly, or not at all. this can be expanded to distrusts and friends-of-friends/friends-of-foes etc etc, but only long as it remains private and simple, with no outward "ranking" of users in any way. &mdash; Asbestos | Talk  16:40, 12 July 2005 (UTC)

Individual ratings by individual people should be hidden.

Each editor sets an individual trust table. Results are tallied into the main trust table that anyone can look at. There should be different categories such as Accuracy, NPOV, Fairness in discussion. People may have a high trust rating in one category and not in another. (But what about erractic people, highly trustworthy in some categories of knowledge and totally untrustworthy in others?)

Perhaps ratings should be positive and negative, with a null setting also to indicate "no rating" which is not the same as trust/distrust rating of 0. Or perhaps there should be 2 tables, a trust table and a disrust table as a way to indicate those who are erratic or whose edits are controverial (which is not necessarily bad). I would expect sometimes to find the same editor to be both highly trusted and highly distrusted. Let's see that kind of thing. The two tables could be used together to create a controversy rating.

As to users trying to regain trust, they don't have to know who distrusts them. In many cases they will probably be quite able to figure it out. Who have they been arguing with? Who has been correcting their old work? But the goal is to gain trust generally from everyone. If one's work is found to be untrustworthy by subsequent editors, then go back and clean stuff up yourself and be more careful in the future. Or continue to be first-draft creators if that's what you like doing.

Of course some people will play the game wrongly, purposely seeking out trusted people and showing-off trustworthiness by agreeing with them in an argument. One could partly neutralize this by allowing the results of the main table to be displayed in different ways using different algorithms to see the differences in trust rating in particular categories when ratings by highly trusted raters are valued more highly and when they are not. jallan 18:02, 6 Jul 2004 (UTC)

awesome idea
Can you put me on the list as trusting User:Christopher Mahan, User:Kim Bruning, User:Mark Richards, User:Eloquence and User:MyRedDice. Oh and I both trust User:Jimbo and trust the people who he trusts ;). I assume I'm not sposed to edit the list myself, from the above discussion. Looks hard to edit anyhow. Thanks, Sam [Spade] 20:57, 6 Jul 2004 (UTC)


 * I think ideas in this vein may make all the difference towards reaching 1.0 quality levels. As an aside, my advice is to allow only contributing members to be ranked/rank others thusly (i.e. charge a minimum donation from those who take part. A long term idea, obviously, but it solves certain other issues, and would generate funds.). Sam [Spade] 21:00, 6 Jul 2004 (UTC)

OK Idea
But to make this work properly, it would probably need a real database behind it with the ability to alter and rate your degree of trust etc, the ability to conceal who you trust and distrust from other users, etc. Sjc 04:56, 14 Jul 2004 (UTC)

How to Game this System

 * Hey, buddy. Yeah, you. I noticed that you're looking bad on the Web O' Trust.
 * How would you like a "trust boost" from me?
 * What do you have to do? Well, it's simple: there are these articles listed on VfD that I want to see {kept|deleted}. If you'll vote my way, I'll hit you up with some good old trust.

(later)
 * Hey, now that I've upped your trust in exchange for several votes, it looks like we're kinda aligned. Tell you what, I'm starting this thing, I call it the "Democratic-Republican Party". The trick is, we all vote together on VfD and we all up each other's trust level, and we all revoke trust from our opponents (I call them the Tories). Then we'll run Wikipedia!

(later)
 * Yeah, well all of us in the "Red User Pages" trust other "Red Pagers", and we trust few or none of the "Blue Pagers". And the "Blue Pagers", well they trust few or none of us. But I think we have over 270 votes on VfD for {dubious page full of the sort of POV preferred by "Red Pagers"} to their 260, and on the poll question, "WikiPATRIOT Act to allow immediate perma-ban of Blue Pager Evildoer Terrorsists".

(later)
 * We can't stay here always fighting the Blue Pagers over every little thing. And they're recruiting new users to share "trust points" faster than we are. Look, the Blue Pagers won't fight back. Once we delete Fort Sumter, they'll have to agree to let us take our half of the database and have our own wikipedia. What can do wrong?

-- orthogonal 13:47, 14 Jul 2004 (UTC)

Vertrauensbildende Massnahmen (Confidence-building measures)
The German Wikipedia are trialling a trust system where people link to a user's subpage if they trust them. By clicking "what links here" on a user's "trust page", you can see who trusts them. See de:Benutzer:Elian/Vertrauen for an example and de:Wikipedia:Vertrauensnetz for a decription in German. Angela. 12:37, Jul 24, 2004 (UTC)

Trust Metric
Amen on looking at Advogato. Seems to me like the three-rank system and peer-rating they use works fairly well. This could also be used on #wikipedia IRC to divy out channel op status and promote harder work and better behavior all around.
 * Quinobi 03:12, 21 Sep 2004 (UTC)
 * I retract the third sentance of that statement. What was I thinking? Quinobi 23:27, 15 July 2005 (UTC)

Typos?
The top of the page says "Here users can list who they trust to make good edits (first column) and also list those who they trust to trust good editors. In the future this table might be used to create bespoke recent changes where they users you trust (directly or indirectly) are missed out, or listed in a light font)."
 * 1) But the table has 4 columns now, and none appear to be "those who they trust to trust good editors".
 * 2) What's "they trust to trust good editors"? Was it supposed to be "...to be good editors"?
 * 3) *"trust to trust good editors" = "Trusts the people whom the following trust" -- Jmabel | Talk 19:32, July 15, 2005 (UTC)
 * 4) "bespoke recent changes" -- I assume this means "custom-taiored recent changes lists"
 * 5) * Precisely. Not a word I'd have chosen, but clear in its meaning. -- Jmabel | Talk 19:32, July 15, 2005 (UTC)
 * 6) What's "missed out"?
 * 7) whould't you want changes by people you trust highlighted, not grayed out?
 * 8) * No, the idea is to be able to check the untrusted edits. -- Jmabel | Talk 19:32, July 15, 2005 (UTC)
 * 9) **To clarify: the main reason that people check the recent changes list is to spot vandals as they're vandalizing (see RC patrol). This proposal would basically have a similar effect as the button currently at the top of the RC list, "Hide logged in users", but would allow finer shades of trust (rather than just logged in/not logged in). &mdash; Asbestos | Talk  22:49, 15 July 2005 (UTC)

I'd fix these myself, but I don't know what they're trying to say... Steve Summit 18:32, 14 July 2005 (UTC)

Alternative Concept: Quantity of Editors
It seems like the trust concept has two major limitations: (1) it discourages new users as their input is less valued and (2) it's only valuable to those who have put together a trust network.

Here's my alternative idea: If Wikipedia is based on the theory that more editors creates a better product, shouldn't the trustworthiness of a page be based on the editor diversity. For example, if one editor wrote 99% of the copy, this would be much less trustworthy than an article that was writen by 25 different people with no one person contributing over 10%. SanDiegoPolitico 08:28, 30 December 2005 (UTC)


 * Nope. This would put a high trustworthiness on some of the most controversial pages, full of unreferenced POV statements, subject to mass warring. -- Jmabel | Talk 19:52, 30 December 2005 (UTC)

Good point. I wonder if there's a way to mathematically differentiate between a page that's being changed as part of a healthy evolution and being changed as part of a POV war. Maybe using the rention of past content as part of the measure (exempting rollbacks). It's beyond my technical skill set but it would be interesting to run a few statistical reports on a selection of articles & see if any patterns appears. -- SanDiegoPolitico 18:22, 31 December 2005 (UTC)