Trust List Analysis - Alt Account Identification

[dogie]

It does however not - assuming here, please correct me if Im wrong - take into account how likely it is for someone to end up on a given trust list in the first place. From the trust.txt I downloaded just a few minutes ago I see that - as expected - certain users are more often on someones list than others. You have a top list for this as well IIRC in your other thread. E.g. theymos is on 170 different trust lists, dogie on 34.

It does not, that's why I'm manually screening anything I release at the moment, there are plenty of "strong correlation" results I filtered out because they're common results. Depending on how much time I invest in this, I may filter out some of the top people altogether.

Which could be argued to bring in bias since it relies on your perspective and past experiences.

If 2 people both have Theymos, John K and BadBear in their lists, the correlation would be 100% but it doesn't tell us anything about the relationship between the two accounts because its so common. The current version doesn't weight based on frequency yet so it needs manual filtering.

There are different ways to get the same result and only the direct copy would be noticable by your reasearch. Additionally for those that actively try to hide their alts its very easy to manipulate their list in such a way that they no longer end up over 0.8 or any other limit for that matter.

The more you ask about the algorithm the easier it is to game, but as you'll see in later releases its no where near as easy as you're making it out to be to mask your list if you want it to be functional. The last batch I processed last night used ratings as low as 0.5 and there are plenty of strong correlations down there, so a future run will take even more. I also set maximum matches to 5 thinking most of these guys wouldn't have more than 5 but there are plenty already hitting that boundary.

[1713588056]

1713588056

[1713588056]

1713588056

[1713588056]

1713588056

[1713588056]

1713588056

[Quickseller]

It did not for Quickseller and the (now known) alt, just to name one example of someone with a hidden alt and a highly modified trust list.

There are different ways to get the same result and only the direct copy would be noticable by your reasearch. Additionally for those that actively try to hide their alts its very easy to manipulate their list in such a way that they no longer end up over 0.8 or any other limit for that matter.

I don't see any reason to use more then one trust list across multiple accounts, at least no legit reason. (one possible reason would be to make someone look more trustworthy or less trustworthy then they really are). If I want to see someone's trust score using QS's trust list then I will look at their profile from QS, and if I want to see their trust score with DefaultTrust then I will look at their profile from an alt.

The only time that I would modify the trust list of an alt account would be for testing purposes to look at someone's trust list/trust network. This is now somewhat of a moot reason because the entire trust network is published every Friday night (EST, Saturday morning GMT).

This kind of research could be used to monitor for scammers who are attempting to farm trust, as many trust farmers tend to ask for both trust feedback and to be added to other's trust lists.

[shorena]

-snip-
Which could be argued to bring in bias since it relies on your perspective and past experiences.

If 2 people both have Theymos, John K and BadBear in their lists, the correlation would be 100% but it doesn't tell us anything about the relationship between the two accounts because its so common. The current version doesn't weight based on frequency yet so it needs manual filtering.

I agree that some additions to a trust list might be common while others are not, my point was that it brings bias if you alone decide it what is common and what not.

There are different ways to get the same result and only the direct copy would be noticable by your reasearch. Additionally for those that actively try to hide their alts its very easy to manipulate their list in such a way that they no longer end up over 0.8 or any other limit for that matter.

The more you ask about the algorithm the easier it is to game, but as you'll see in later releases its no where near as easy as you're making it out to be to mask your list if you want it to be functional. The last batch I processed last night used ratings as low as 0.5 and there are plenty of strong correlations down there, so a future run will take even more. I also set maximum matches to 5 thinking most of these guys wouldn't have more than 5 but there are plenty already hitting that boundary.

What if I told you that I did not post this yesterday because I was pondering whether or not I should post how to easily game your analysis, but still see identical trust ratings? I have not gambled for a while, would you be willing to make it a small bet?

[dogie]

-snip-
Which could be argued to bring in bias since it relies on your perspective and past experiences.

If 2 people both have Theymos, John K and BadBear in their lists, the correlation would be 100% but it doesn't tell us anything about the relationship between the two accounts because its so common. The current version doesn't weight based on frequency yet so it needs manual filtering.

I agree that some additions to a trust list might be common while others are not, my point was that it brings bias if you alone decide it what is common and what not.

The alternative is posting uninterpreted data which pretty much suggests everyone = everyone to some degree, which will include things like BadBear = everyone, Theymos = everyone etc etc which is pointless. That's only going to be used as a weapon to attack people. I post the trust lists (truncated to I think 160 characters) alongside any potential matches so you yourself can decide if you think its a match or not.

There are different ways to get the same result and only the direct copy would be noticable by your reasearch. Additionally for those that actively try to hide their alts its very easy to manipulate their list in such a way that they no longer end up over 0.8 or any other limit for that matter.

The more you ask about the algorithm the easier it is to game, but as you'll see in later releases its no where near as easy as you're making it out to be to mask your list if you want it to be functional. The last batch I processed last night used ratings as low as 0.5 and there are plenty of strong correlations down there, so a future run will take even more. I also set maximum matches to 5 thinking most of these guys wouldn't have more than 5 but there are plenty already hitting that boundary.

What if I told you that I did not post this yesterday because I was pondering whether or not I should post how to easily game your analysis, but still see identical trust ratings? I have not gambled for a while, would you be willing to make it a small bet?

I don't really understand your point. This analysis isn't meant to be the "no one can have an invisible ult account ever again" system, its simply used to unearth several 100 of them already in existence. Most of which are used for evil purposes.

[shorena]

-snip-

-snip-
What if I told you that I did not post this yesterday because I was pondering whether or not I should post how to easily game your analysis, but still see identical trust ratings? I have not gambled for a while, would you be willing to make it a small bet?

I don't really understand your point. This analysis isn't meant to be the "no one can have an invisible ult account ever again" system, its simply used to unearth several 100 of them already in existence. Most of which are used for evil purposes.

Maybe I just expected more than similarities between trust lists. Its certainly interesting work, I just dont think it can do what you suggest. On the other hand I might not have the background to judge the significance.

[dogie]

-snip-

-snip-
What if I told you that I did not post this yesterday because I was pondering whether or not I should post how to easily game your analysis, but still see identical trust ratings? I have not gambled for a while, would you be willing to make it a small bet?

I don't really understand your point. This analysis isn't meant to be the "no one can have an invisible ult account ever again" system, its simply used to unearth several 100 of them already in existence. Most of which are used for evil purposes.

Maybe I just expected more than similarities between trust lists. Its certainly interesting work, I just dont think it can do what you suggest. On the other hand I might not have the background to judge the significance.

Its an analysis of trust lists, its not magical That being said if I had access to other forum data I'm sure it could get exponentially more accurate. Have a look at some of the batch 2's I've released, lots of alts.

[dogie]

Added batch 2, including some lower tier results. These should generally be ignored in isolation although can often identify some outliers to an existing alt network. Also added a visualisation of the first 2 batches of results.

[Blazr]

bbit	hashie	1	0.6	neg_Vod&neg_Blazr	neg_Vod&neg_Blazr

√ Though bbit did not initially belong to hashie. It was bought.

Hmm I dont know about that... I don't think bbit was bought. I was the one who outed hashie originally and most evidence I've seen points to hashie and bbit being two separate individuals.

Glancing through this list I see a few other accounts that I know for sure aren't alts. Nice idea but it isn't very accurate at all. Maybe you aren't thinking about this the right way, take a look at this one:

ahmedbodi

VERUMinNUMERIS

0.56

1.95

ahmed_bodi&mrbodi

markm&frobley&kelsey&ahmed_bodi&deadsea33&Vlad2Vlad&Dexter44&cinnamon_carter&… er&IXC2XIC&mrcashking&SolomonRising

I would think that ahmed_bodi and mrbodi would more likely be alts of ahmedbodi rather than VERUMinNUMERIS.

[dogie]

Hmm I dont know about that... I don't think bbit was bought. I was the one who outed hashie originally and most evidence I've seen points to hashie and bbit being two separate individuals.

If they are 2 separate individuals then they are associated, which is what this analysis highlights.

Glancing through this list I see a few other accounts that I know for sure aren't alts. Nice idea but it isn't very accurate at all.

Glancing through the list will of course highlight names you say aren't alts, that doesn't mean the other 97% aren't associated in some way.

Maybe you aren't thinking about this the right way, take a look at this one:
I would think that ahmed_bodi and mrbodi would more likely be alts of ahmedbodi rather than VERUMinNUMERIS.

You're looking in the really low association table which doesn't mean much. The higher association table already highlighted that ahmedbody = ahmed_bodi, and mrbodi isn't in the data because it has an unedited trust list. VERUMinNUMERIS is already strongly associated with 3 other accounts so its more than possible its the same person controlling them all.

So in conclusion the analysis identified: 2 alt accounts with likely the same owner, 4 alt accounts with likely the same owner, and potentially a link between the two sets. I'd call that a success (see the diagram).

[monkeynuts]

There is a danger here, that this analysis is identifying 'communities' within bitcoin rather than alts

for example - TookDK / Wheresmycoin

Both I know well. Both I have bought and sold physical crypto coins from and to. Live on opposite sides of the world. Not alts.

But are both keen physical crypto currency buyers / sellers / collectors. One example of a very niche community who buy / sell and trade with fellow members of the same niche community, that will by inference result in very common trust profiles.

Reckon this is stating that the physical crypto bods need their own section. I guess there will be other similar examples in the lists.

[dogie]

There is a danger here, that this analysis is identifying 'communities' within bitcoin rather than alts

It may pick up trading communities, yes.

for example - TookDK / Wheresmycoin

That's just an error, I flagged them wrong at some point when comparing 1000s of lists. They probably wouldn't even qualify for the lower tier of association.

[Lesbian Cow]

There is a danger here, that this analysis is identifying 'communities' within bitcoin rather than alts

for example - TookDK / Wheresmycoin

Both I know well. Both I have bought and sold physical crypto coins from and to. Live on opposite sides of the world. Not alts.

But are both keen physical crypto currency buyers / sellers / collectors. One example of a very niche community who buy / sell and trade with fellow members of the same niche community, that will by inference result in very common trust profiles.

Reckon this is stating that the physical crypto bods need their own section. I guess there will be other similar examples in the lists.

CJBianco Lesbian Cow 0.61 6.02


I do not have any alts.  Both of us do trade physical bitcoins - it is a fairly small community so it is not surprising we have similar trust lists.

[TookDk]

There is a danger here, that this analysis is identifying 'communities' within bitcoin rather than alts

It may pick up trading communities, yes.

for example - TookDK / Wheresmycoin

That's just an error, I flagged them wrong at some point when comparing 1000s of lists. They probably wouldn't even qualify for the lower tier of association.

Yes, your algorithm got a false-positive on this one.
Wheresmycoin is located 10000 Km from me, literately on the other side of the globe.

[Quickseller]

There is a danger here, that this analysis is identifying 'communities' within bitcoin rather than alts

for example - TookDK / Wheresmycoin

Both I know well. Both I have bought and sold physical crypto coins from and to. Live on opposite sides of the world. Not alts.

But are both keen physical crypto currency buyers / sellers / collectors. One example of a very niche community who buy / sell and trade with fellow members of the same niche community, that will by inference result in very common trust profiles.

Reckon this is stating that the physical crypto bods need their own section. I guess there will be other similar examples in the lists.

CJBianco Lesbian Cow 0.61 6.02


I do not have any alts.  Both of us do trade physical bitcoins - it is a fairly small community so it is not surprising we have similar trust lists.

And the drrama starts

I would vouch that this is a true statement, at least the part about CJBianco and Lesbian Cow being different people.

I would not be surprised if many people in trading communities end up being shown as potentially the same person. 

[notlist3d]

Very nice work I really like the idea.  And the method you used seems pretty sound, as some said a few false positives but I suspect you found a good  amount of alt trust with this.

What tool did you use to do this?  Did you make it yourself? Can you tell us some about it.  I enjoy learning how things work.

[bitcoin revo]

I would not be surprised if many people in trading communities end up being shown as potentially the same person. 

Well, the tool shouldn't really be used as anything more than an extra bit of evidence, IMHO. There are lots of possibilities that the "alts" were separate people who just had similar trusting personalities.

[Mitchell]

I would be very surprised if CJBianco was an actual alt of Lesbian Cow (or vice versa).

[dogie]

Very nice work I really like the idea.  And the method you used seems pretty sound, as some said a few false positives but I suspect you found a good  amount of alt trust with this.

What tool did you use to do this?  Did you make it yourself? Can you tell us some about it.  I enjoy learning how things work.

Most things are in excel, yes I made it, no can't say too much about it but its based off the trust list data set Theymos published.

[monkeynuts]

Very nice work I really like the idea.  And the method you used seems pretty sound, as some said a few false positives but I suspect you found a good  amount of alt trust with this.

What tool did you use to do this?  Did you make it yourself? Can you tell us some about it.  I enjoy learning how things work.

Most things are in excel, yes I made it, no can't say too much about it but its based off the trust list data set Theymos published.

Can the outputs of your work be a feed into the work tspacepilot did ? ("Statistical Methods for Natural Language Processing"). Trust list identifies possible alt candidates, language processing gives probability ?

[dogie]

Very nice work I really like the idea.  And the method you used seems pretty sound, as some said a few false positives but I suspect you found a good  amount of alt trust with this.

What tool did you use to do this?  Did you make it yourself? Can you tell us some about it.  I enjoy learning how things work.

Most things are in excel, yes I made it, no can't say too much about it but its based off the trust list data set Theymos published.

Can the outputs of your work be a feed into the work tspacepilot did ? ("Statistical Methods for Natural Language Processing"). Trust list identifies possible alt candidates, language processing gives probability ?

Yes. This really needs more data sources but they'll most likely have to come from Theymos. If you think this is hit and miss, language processing is 10x worse and would probably match everyone to everyone else. There is also no way to override or manually check it by human eyes as all it gives you is a name and a number.