People ask me all the time, “Who can I trust who won’t screw me over?” or “How do I know I can trust you?”
As to myself, I could point to things like:
- The BBB using my research here
- Or the National Consumers League referencing WorkAtHomeTruth here
- Or the FTC letter I have sent to me thanking me for helping them on the In Deep Services case involving one of the top 100 spammers in the world according to the ROSKO database at the time (I spent A LOT Of time helping them on that).
- Or documentation I provided (by request) to another branch of federal law enforcement detailing an extensive criminal network, including over a dozen corporations, LLCs, several hundred websites, and an alarming number of D.B.A.s – all across multiple countries.
- Or the numerous requests for information about home business fraud from places like National Public Radio, ABC’s 20/20, The New York Post, The Sydney Morning Herald
- Being asked to write the chapter on home business fraud by an author who works closely with an assigned federal agent and was the only non-agency person at one of the more recent white collar crime initiative meetings.
- Or the power user spotlight at business.gov
Sounds impressive. But is it? And does it really say anything about how much you can trust me?
Maybe or maybe not.
Who’s to say that one day I didn’t wake up and say, “Hey, now that everyone will trust me, it’s time to screw them all over and collect big $$$! (I DID NOT do that, but it is POSSIBLE for someone to do that – AND I’ve seen it happen on numerous occasions).”
Social Network Analysis and Trusted Hubs
I first learned of the concept of Trusted Hubs when studying Search Engine Optimization. However, a video from SEOMoz about TrustRank really helped me understand how useful the concept could be – even outside of the field of link-building and SEO.
Here’s the video from SEOMoz about TrustRank and LinkBuilding:
It appears that he is basing much of his discussion on the type of research in this paper about Combating Spam With TrustRank.
Social Network Analysis and “Trusted Hubs”
So can anything similar to TrustRank be used when it comes to evaluating reputation of a company or person selling “Internet Marketing Training, SEO Training, or providing other types of product or service?
One of the more interesting papers I’ve read on this possibility is the paper “Reputation and Social Network Analysis in MultiAgent Systems” by Jordi Sabater and Carles Sierra which starts with the statement (Abstract):
“The use of previous direct interactions is probably the best way to calculate a reputation but, unfortunately this information is not always available. This is especially true in large multi-agent systems where interaction is scarce. In this paper we present a reputation system that takes advantage, among other things, of social relations between agents to overcome this problem.”
On important thing to keep in mind that “reputation” does not necessarily equal “trustworthiness” (although it might)
I highly recommend going through the paper as it provides a lot of great insight into the pros, cons, and problems associated with using social network analysis as a means to calculate reputation. Also, there are important distinctions made in the paper that would be too detailed to go into here.
A Few Takeaways From Sabater and Sierra’s Paper as Applied to the Realm Of Internet Marketing Training
Where “wi” is the “witness” providing information to “a” about “b”, then:
“socialTrust(a,wi,b) (is defined) as the trust degree that agent a assigns to wi when wi is giving information about
b and taking into account the social relations among a, wi and b.”
And also, one of the more basic, but still important, ideas from the paper is that:
where “wi” is the “witness” providing information to “a” about “b”, then:
“IF coop(wi; b) is high
THEN socialTrust(a;wi; b) is very bad”
that is, if the level of cooperation between wi and b is high then the trust from the point of view of a on the information coming from wi related to b is very bad. The heuristic behind this rule is that a cooperative relation implies some degree of complicity between the agents that share this relation so the information coming from one about the other is probably biased.”
A recent example of this can be seen in this post at the SaltyDroid.info site about “The Internet Marketing Syndicate”.
Frank Kern’s “Why You Must Form A Syndicate” video in light of insights from Jordi Sabater and Carles Sierra’s paper about using Social Network Analysis to attempt to analyze reputation
In Frank Kern’s “Why You Must Form A Syndicate” video, Frank Kern advises businesses/people to form a “Syndicate” within their market and:
- Work together with competitors to constantly be delivering “value” to customers
- Work with a maximum of 10 people/businesses who are your biggest competitors
- Reach out to the biggest competitors by setting up a meeting.
- Develop a group of people/businesses (Syndicate) that promote each other, DELIBERATELY endorse each others products, and grow each others’ businesess.
- Aim for a level of cooperation that would occur naturally if the fear of competition wasn’t in the way.
- You have to deliberately cultivate, create, and position yourself as the leader of a “trade union” (a.k.a. Syndicate a.k.a. the “A Team”) within your market.
Frank Kern also goes on to state:
- All the top people in the internet marketing space promote each other.
- A lot of the top people are in a “Trade Union” (a.k.a. Syndicate – a.k.a. the “A Team”)
- The “B” teams buzz supports the “A” teams buzz through the social proof mechanism which amplifies the “A Team”‘s message.
What Might Social Network Analysis Say About The Value Of Product Recommendations And Value Of The Products (Price) Coming From The Internet Marketing Syndicate?
Again, remember the formula where “wi” is the “witness” providing information to “a” about “b”, then:
“IF coop(wi; b) is high
THEN socialTrust(a;wi; b) is very bad”
In other words if the assumptions in Jordi Sabater and Carles Sierra’s paper are correct, then it’s likely that any information ABOUT products from Syndicate members OR the value placed on products (i.e. price) could quite possibly be “very bad”.
10/2/2010 correction: It looks like I made a pretty serious error in the statement above as what they are talking about is the “perceived value” of the witnesses information.
“In other words if the assumptions in Jordi Sabater and Carles Sierra’s paper are correct, then it’s likely that any information ABOUT products from Syndicate members OR the value placed on products (i.e. price) could quite possibly be [perceived as] “very bad”.
That doesn’t throw out the entire concept of this post, but it’s significant. I go into that more in a comment at the bottom of this post.
End 10/2/2010 correction
Does The Internet Marketing Syndicate Truly Believe They Are Delivering “Value”
Another question that comes up in the video is whether or not the members of the Internet Marketing Syndicate believe they are delivering value. The Salty Droid seems to imply that he doesn’t think the Internet Marketing Syndicate believes they are delivering value when he displays the word value within quotation marks in the video.
Of course if it’s true (as has been claimed by some) that members of The Internet Marketing Syndicate promote each others product no matter what, then of course it’s less likely (although still possible) that they truly believe they are delivering value to their prospects and customers than if they only promote the product if they believe it is beneficial to their subscribers/clients.
Sabater and Sierra make an interesting distinction in this area when they say make a distinction between “sincere” information (information that an agent believes is true) and “true” information (information that actually is true).
Why “Social Proof” Is Usually NO Proof At All
I’ve heard a lot of people mocking the idea of “Social Proof” saying that it is meaningless and not REAL proof.
But the field of Social Network Analysis provides a compelling REASON why “Social Proof” is usually NO proof at all:
Sabater and Sierra state:
“the information that comes from other agents can be correlated (what is called the correlated evidence problem). This happens when the opinions of di fferent witnesses are based on the same event(s) or when there is a considerable amount of shared information that tends to unify the witnesses’ way of thinking”. In both cases, the trust on the information shouldn’t be as high as the number of similar opinions may suggest.
The Problem Of “Perceived Neighborhood Reputation”
Jason Fladien hinted at this in his recent post called “Stay Away From These Products” in which he states, “What I have just done is probably pissed off a lot of my so-called peers. Lost a lot of potential joint venture partners. Alienated alliances that could bring me in big bucks.”
In other words when he talks of his “so-called peers” he seems to be hinting at the possibility that many people will accidently lump him in with the more questionable internet marketing trainers merely because he does training in that field as well.
The question then becomes how can you distance yourself from a “Neighborhood Reputation” that doesn’t reflect who you are and doesn’t reflect your values.
The Temporal Aspect Of Reputations
Sabater and Sierra make a point to clarify that, “reputations also have a temporal aspect (the reputation value of an agent varies along time)”.
In other words opinions can change, relationships can change, and a person’s values can change.
And like it or not, the important questions to be asked (especially early on) are:
- Is the change real?
- What is the motivation behind the change?
- Why didn’t the change occur earlier?
Possible Solutions To “The Hidden” Information Problem In Social Network Analysis
One problem inherent in social network analysis has to do with whether the true relationships between and among people are really known.
For example, in a competitive environment a person/business has a pretty big motive not to disclosure their relationships.
However, in certain scenarios this isn’t as much of an issue.
A good example of this is this case study about a Slumlord Conspiracy from Orgnet (Uncloaking a Slumlord Conspiracy with Social Network Analysis) which involves a scenario where it is possible to gain data about relationships that is more likely to be accurate than in other scenarios where relationship information is dependent upon the people/businesses involved in the social relationships who may deliberately hide information OR not know the information.
The abundance of free and paid tools, such as those I list in the section on Competitive Intelligence Tools & Techniques make it possible in many situations to get potentially quite accurate data and information – although only potentially, because as discussed in Jennifer Xu’s and Hsun chun Chen’s Criminal Network Analysis and Visualization, there’s always a chance the information is:
- Incomplete
- Incorrect (intentially or unintentially)
- Inconsistent
…amongst other potential data accuracy problems discussed in their paper.
However, if you can join the “harder data” that can be revealed through documentation with the “softer data” about what people/businesses self-disclose about their relationships then it should be possible to get a better idea of the trustworthiness of different nodes and neighborhoods in a social network.
Related Research And Information:
- Computational Trust And Reputation Models
- Criminal Network Analysis and Visualization
- Artificial Intelligence (AI) and systems that reason plausibly from uncertain evidence
Related Posts:
- Recent News Of Interest (Especially the part about Recorded Future, technology in which the CIA and Google recently invested.
- The Internet Marketing Syndicate
- Is Internet Marketing “Fixed”? (Part 1)
- Stay Away From These Products
- When Is Viral Marketing Very Very Bad?
If You're Struggling To Make Money Online - Click Here To Watch This Free Video And FINALLY Get Answers To All Of Your Questions About Making Money Online
Paul, this is a very ambitious post. Here are some of my thoughts.
It used to be that we could identify fraud in the classifieds business ads because the same ad would appear in many different geographical locations – and each ad referring to a unique local opportunity, for example a local vending route. Clearly, this was an impossibility.
Only reading one or two newspapers would not reveal this pattern – but looking at the online classifieds would.
Of course, anyone looking for a business opportunity would not be sufficiently skeptical of an ad to see if it was repeated. (Just thinking about how much those ads cost would alert you to the fact these people had no money to run or support your business.)
But this detection device or tool only lasted 5-6 years, as fraud moved away from the classifieds into the internet proper.
However, the basic technique still works, yet has the same limitations. Yes, we can probably detect similarity in pitches and so discount them; but no, the people who buy the pitch are not sufficiently skeptical to be looking for similarly sourced information.
Can some of these tools assist in uncovering patterns of fraud and so preventing fraud? I am not sure. Frank Kern’s syndicates are just groups of people getting together to bombard people with different variations of the same pitch: even if I know that, might I not get tired after awhile and simply buy one of the products to shut off the channel?
Also, I have some trouble with the conceptual identification of “spam” = “fraud”. Even if we could identify trusted sites, and then it turned out that 3 or 4 degrees away from these trusted website, we are likely to find spam, I am not sure how this helps.
For example, much of the FTC website would on this methodology be a trusted site – yet, I wouldn’t rely upon a single thing they said about how to really avoid fraud. It is largely pap generated by a regulator who has no idea about how to stop the flow of fraud.
I may have misunderstood the overall tenor of your remarks, and invite corrections by you or other readers.
I’m going to have to think through this again when I have a bit more time in a day or so.
One thing I should clarify is I personally think SOME of the information from people who might be (I don’t know for a fact) members of the so-called “Syndicate” is actually quite good – especially Jeff Johnson’s traffic information (and I don’t agree that they are always variations of the same pitch, although sometimes they are).
But let’s for the sake of example say there are other members of the group who are somewhat (or even very) shady.
I don’t think “guilt by association” is reasonable if you don’t know the STRENGTH of the connections between people within the group. So, one thing I’m interested in is how to analyze the real strength of connections within something like a closed group of affiliate marketers or JV partners.
Because of the nature of the internet and the amount of relationship data and information people often disclose (directly through public statements or indirectly through observing the types of interactions they have with other members of a group – i.e. do A & B ALWAYS recommend each others products, or does A sometimes NOT recommend B’s product), I think something like this should be possible.
Ironically, I have a real issue with the use of the word “fraud” (mainly because I am skeptical about “proof” as explained in the next paragraph). As it’s so emotionally charged and why I am thinking that possibly looking at networks from the perspective of trust and reputation could possibly be more useful from the standpoint of potential consumer as to who they can likely trust (I only started thinking about this).
I may be skeptical about “proof” to a fault. For example, even with all the connections I make using technical analysis of website properties, domain ownership, shared analytic codes, hosting history, corporate ownership, related D.B.A.s, shared nameservers, mail servers, etc. – I am HUGELY skeptical that I’ve ever “proven” anything and I am VERY reluctant to call something “fraud”.
In fact, it’s when it dawned on me why the concept of perjury is SO important in the legal system.
I hear what you’re saying about the FTC (and other so-called consumer protection agencies) and I’ll try to say more about that when I post again here.
Re proof of fraud:
Paul, the legal use of “proof” in a civil case simply amounts to this: is A more reasonable the B, where A is put forward by the plaintiff and B by the defendant.
Unlike science, or the Scottish criminal system, there is no option of “neither”.
Interesting. Sounds like I should head over to Scotland.
Just was checking my RSS reader and noticed one of the blogs I’m subscribed to has good information about sentiment analysis:
http://lifeanalytics.blogspot.com/search/label/sentiment%20analysis
which also could be useful.
Michael,
I’ll give a very blatant example of what I’m thinking just for the sake of clarity.
What I haven’t seen as much of (although it may very well be out there) is combining concepts from criminal network analysis (or competitive intelligence) which tend to focus on hard data with social network analysis which focuses on somewhat softer data (it seems anyhow).
Let’s say that 3 people (Person A, Person B, Person C) form a JV group that always agree to promote each other no matter what.
And let’s say there is hard data such as numerous State and Federal Law Enforcement actions individually Person A, Person B, and Person C. Or maybe those actions are only against Person A & B, and person C is listed in the ROSKO top 100 spammers database. In any event there’s external hard data about each person.
Then let’s say persons D & E joins the JV circle and the only thing I know about persons D & E are:
D states publicly (on their blog, say) that consider A, B, & C good friends – or let’s say D makes numerous tweets from Twitter complimenting A, B, & C, and recommending their products.
Person E’s relationship to A, B, & C is unknown.
In this deliberately blatant example then at this point D would likely have a much lower perceived reputation (or possibly trust) level than person E.
So the key is having a system for identifying strength of relationships which is something that Social Network Analysis seems to focus well on.
In this example I made it really obvious how strong Person D’s perceived relationship to A,B, & C is, but because of the data mining capabilities online, this kind of thing is already being done by several companies.
Again, what I haven’t seen as much of (although it may very well be out there) is combining concepts from criminal network analysis (or competitive intelligence) which tend to focus on hard data with social network analysis which focuses on somewhat softer data (it seems anyhow).
Yes, that helps.
But generally in a criminal investigation, one is trying roll up the small fry so they will blab.
With internet crime, it is very unclear as to who the small fry are and who are the real promoters. (I am thinking of the clearly fraudulent rebill programs.)
I am not sure that we are ever going to see a footprint on the internet from many of these criminal promoters. (Most of the biz op fraud guys use aliases and disposable cell phones.)
My sense is that a) yes, you are on to something, but b) it probably has to do with quantifying credit risk rather than uncovering fraud.
Thanks, Michael
I really only started thinking about it last week. I wasn’t aware of the SNA module at the time which seems to also focus more on credit risk.
And I hear what you’re saying about finding the real promoters. It’s always fun to try to solve the unsolvable, though, as it can lead to unexpected useful results.
Here is my standard observation: when an internet promotion suddenly explodes and ranks higher than the regulator for that industry, then it is very likely fraud.
If we could map the social network that caused this, and this was known, we would reduce a certain type of internet fraud, without having to wait for a lot of consumer complaints.
Just came across this December 2009 article about a new SAS module which takes a very similar approach actually:
Fighting Fraud With Social Network Analysis.
Here are 3 key points (as far as how they relate to this post) from the above referenced article:
1) “Introduced by SAS earlier this year, SNA software gathers data from multiple sources, links individuals who share key pieces of information or engage in transactions with each other and then presents the associations using “a unique network visualization interface.”"
2) “”We all have social networks and they are generally benign social networks, but if you lock onto somebody who is an Internet criminal and you have enough data at your disposal, you can actually link out their social network,” he said.”
3) “Predictive analytics can be used to determine the likelihood of an individual becoming a bad customer and whether the individual is too much of a risk to even begin conducting business with, Gill pointed out.”
And apparently the software has the capacity to handle potential problems such as false positives, etc.
“One offensive strategy organizations can take, according to Swecker, is to inject some uncertainty into the sale of customer data. One of the biggest vulnerabilities cybercriminals have is trust in each other in their carding sites, he said.
Cybercriminals rely on trust, their relationships and the names they use on the Internet, said Swecker. “If you can undermine that and create uncertainty in their own black markets, I see that as a vulnerability,” he said.”
I agree with this observation entirely – disrupt, disturb and undermine. You are trying minimize the effects of fraud and by reducing its ability to scale, you win.
This may be stretch, but minimizing the ability to scale in this way seems somewhat (albeit remotely) related to your observation at the end of your post here:
http://bizop.ca/blog1/the_franchise_news/2006/11/when-is-viral-m.html
Where you state:
“To effectively combat pyramid schemes we need more “Poison Parasite” advertising. We do not need educate individuals about the mathematics of pyramid schemes because although the math is correct, math challenged consumers are not problem. Unchecked deceit is the problem.”
In both instances your getting into the “bloodstream” of scalability.
It was also the reason I was trying to convince the FTC to focus on the affiliate networks instead of the vendors, since affiliate networks is exactly how the rebills scaled so much. They weren’t interested – or certainly didn’t appear to be.
Of course, what happened already to an extent was that private affiliate networks were set up, so I’m not even sure attacking the problem at the affiliate network level would do anything, although I would think it would have some reasonable impact on decreasing the problem…
…especially since Pascal’s Wager is such a powerful “drug” and there’s a seemingly endless supply of people looking for that “fix”.
After reflecting on this, I believe one thing I wrote in the initial post may not be quite accurate:
“In other words if the assumptions in Jordi Sabater and Carles Sierra’s paper are correct, then it’s likely that any information ABOUT products from Syndicate members OR the value placed on products (i.e. price) could quite possibly be “very bad”.
I think it should be:
“In other words if the assumptions in Jordi Sabater and Carles Sierra’s paper are correct, then it’s likely that any information ABOUT products from Syndicate members OR the value placed on products (i.e. price) could quite possibly be [perceived by a potential buyer as potentially being] “very bad”.
With the correction being “[perceived by a potential buyer as potentially being]”
That’s a pretty significant difference.
I don’t think it destroys the value of the concept behind this post completely, but as I’m thinking it through there are probably many more additional factors that come into play.
Also, there’s probably a need to separate “value” (good or bad, etc.) from how that “value” is marketed and delivered.
Finally, I think there’s a lot to be learned by looking at where the strength/nature of relationship can predict the value of a product delivered vs where it can’t.
The easy answer would be “it can’t”. My GUESS (just a gut feeling) – based on analyzing some of the close-knit group of marketers who are all associated with two prolific spammers – is that there are some cases where it CAN, although any time you get into the area of “value” things can get messy fast.