Aggregation, Not Algorithms, Is The Key To Establishing Trust Online
Image by opensourceway on Flickr.
The biggest issue the new sharing economy is currently facing is one of trust. After issues like the Airbnb debacle and high-end car sharing service HiGear’s closure due to theft, it's clear that trust issues are the main barrier preventing sharing from becoming mainstream.
The answer that everyone seems to be looking for is a some sort of global reputation system. There has been a lot of talk about a single "reputation score" that you could take with you in any new service. At a first glance the idea seems like a winner: with every action you do in the web affecting the score, it would be easy to find the rotten apples.
However, it turns out this idea is fundamentally flawed.
There has been a multitude of attempts to create a global reputation score solution. The most well known recent example is Klout. It measures your "influence" in Twitter and other social networks based on factors like amount of retweets and influential followers, and based on them algorithmically gives you a "Klout score".
The service has captured a reasonably wide userbase and some people take the score very seriously. Some companies have even been reported to be using it in their hiring. This has prompted discussions whether Klout could be the solution for the sharing economy.
However, others have also pointed out Klout's many issues: it is a rather arbitrary number with no clear meaning to most people, its algorithm is not public knowledge, the company can change the algorithm as they please, and it can be gamed by acting "the right way" on Twitter (which includes things like not talking to people with lower scores). Due to these issues, most of people I know seriously question Klout's ability to reliably measure influence.
Image via Socialbrite
And even if the Klout would be the answer in measuring influence, it would not do much good for the sharing economy. The thing with reputation is that it always needs to be in context. If I'm pondering whether a person will bring my power drill back in one piece, their Twitter influence level would be about as useful information to me as knowing that they are a level 53 mage in World of Warcraft.
So why not just use eBay data?
Because of the contextual nature of reputation, there can be no be-all, end-all reputation system. But it can be argued that in the case of sharing economy, the relevant contexts are similar: in most cases, we need to know whether the person in the other end will provide the money, goods or services they promise, or, in cases of Airbnb and other rental marketplaces, return the rented property on time and in good condition.
So should we used use the reputation score from eBay or other major player like that? With technology like OAuth it would be relatively easy for them to offer a method to use their feedback ratings in other services.
This is definitely a step into the right direction, but it's not enough.
Image by opensourceway on Flickr.
I got a chance to meet Airbnb's Brian Chesky during LeWeb and asked him about the trust issue. Obviously they have been thinking about it too, and he said that like eBay, Airbnb could in the future provide a reputation platform that could be used by other services. But then I asked a followup question: does he think that it's problematic that a single company would emerge as the major keeper of people's reputation wallets. He admitted that it's an issue he doesn't have a good solution to.
In believe the solution might be rather obvious: instead of relying on a single existing reputation system, we should rely on many. There should be a single aggregator that would use OAuth to pull all of an individual's transaction-related feedback from eBay, Airbnb, Taskrabbit, Zimride, and other such services to a single "reputation profile". When joining a new sharing service, users could again use OAuth to display their data from this profile. .
The same system could also connect to Facebook to bring another major trust-related component to the table: the friend network. If the reputation profile could tell that you and the stranger wanting to borrow your power drill have three common friends, it would further strengthen your sense of trust.
The Problems with Aggregation
DATABASE at Postmasters, March 2009, by Michael Mandiberg
While in many ways the solution described above sounds ideal, it too has some issues.
For this system to work all the services would need to provide a common method of sharing and using reputation data. I mentioned the issue briefly in my previous posts about the Sharing Economy’s great need for an "API for sharing". So far I didn't find a way to do it in any of the sharing economy sites I checked. But judging from what Brian Chesky said at LeWeb, it's coming, and I'm pretty sure eBay's looking into that direction too.
As soon as the required infrastructure is in place, the next question is: Who is the one responsible for the aggregator? We just discussed the problems related to relying on a single provider. Isn't this situation bringing those problems back on the table?
I think the system could be maintained by a single startup. But there's a caveat: the system needs to be as transparent as possible, which in my opinion means that there should be no complex algorithm involved.
It seems that many are looking for a perfect algorithm to display reputation as accurately as possibly. The problem is, no algorithm is good enough. As soon as one is put in place, people will game it. The more complex it is, the more difficult it is to understand, and the less likely is it to carry any meaning to a regular person. But those who know what they are doing will still manage to fool the system. This is one of the biggest problems with Klout, and it will be just the same with any other service like that.
Image by opensourceway on Flickr.
Shareable has previously covered Trustcloud, a service that is doing something a bit like what I have described. TrustCloud markets themselves as "the trust solution for the sharing economy". They have many similar ideas than described here: you do an OAuth to several other services and pull data from there, and then they provide a "trust cards" for third party services to use.
However, the TrustCloud approach has two major problems. First, they pull their data from Facebook, LinkedIn and Twitter. These services are simply not good enough for measuring reputation. In their blog they criticize Klout, but they themselves are victims of the same issue: they try to measure reputation by using data from a different context. If I have to decide whether I let a stranger in my house to put together my Ikea furniture, it's doesn't bring me much comfort that the person is active on Facebook. It doesn't even tell me anything about whether the person's Facebook friends would actually be willing to have the person staying in their homes or assembling their stuff.
The second issue is that they use an ambiguous algorithm to assign "badges" to people. These badges are related to person's reputation in different contexts, which is a good idea. But how do I know if they are any good at measuring these qualities? Why isn't the algorithm transparent? And can they just one day change it and turn my good reputation into a bad one?
Trust Without Algorithms
"runnerfrog_-_abstracta_103", 3D abstract, computer-generated through genetic algorithms. By runnerfrog on Flickr.
If there are no algorithms, how is reputation measured? My suggestion is to use the method used by the most important online reputation system to date: eBay. In their users' reputation cards, two things are shown: the number of ratings and percentage of positive feedback. That's all the information a trust widget needs to contain. If the user wants more information, they can follow links to the actual profiles in the corresponding services and see details, like textual feedback which often contains valuable information. This way the system is as transparent as possible and cannot be gamed easily or distorted by a third party.
Of course there are still some choices to be made. For instance, TaskRabbit uses a five star rating system. Which reviews are positive? Only those with five stars, or those with something from three to five? But these are decisions best left for each service to decide.
Can the system described in this post be created? It will require some work to get all the big players involved. And what's in it for them? Perhaps the biggest reputation data providers could get part of the revenue that the aggregator makes by offering the solution to smaller sites. This kind of arrangement could also prevent all the parties from overcharging for the data.
I believe this solution would be the ideal approach to the problem of trust in the sharing economy. What do you think?
Comments
Bill Seitz wrote on 01.10.12, 2:53pm:
If each market has an open API, and each user is identified by a shared standard ID, then there doesn't need to be a single aggregator. There can be any number of aggregator websites, or even an aggregator desktop app that pulls from the appropriate services.
The big question is whether any of those services is going to provide their streams for free.
Rebecca Petzel wrote on 01.10.12, 3:04pm:
I think yours is the best proposal I've heard yet! Diversity of inputs, transparency in to the logic of the reputation, and financial incentives that encourage companies to collect and share reputation data.
I feel some urgency around making this happen, as there is definitely incentive for a non-transparent, single entity to get in and win the reputation market. And that, in my opinion, is not a road we want to go down....
Juho Makkonen wrote on 01.10.12, 3:15pm:
Johannes, thanks for your comment. Again, I see that it depends on the context. Frictionless sharing can paint a pretty good picture on what you do online, and of course all that information helps people in some ways to assess your trustworthiness, at least in some contexts. But in the specific case of sharing economy, I think that the most powerful reputation comes from testimonials created by others, and they cannot be achieved by means of frictionless sharing.
But you probably have explored the concept of frictionless sharing a lot more than me, so please share your thoughts on the impact it could make on reputation systems.
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Juho Makkonen wrote on 01.10.12, 3:43pm:
Bill, very true, a single aggregator is not necessary. However, I think there are a few reasons for why a single "middleman" is a good idea.
1) As a user I only have to authorize each data provider service once. If every service has their own system, I have to do it once per data provider per service, and that is too much of a hassle. And there are other usability points too, having a middleman would really simplify things up a lot.
2) There are actually quite a bit of infrastructure to build: for instance, the users might want to have an interface to control which services they trust as data providers, and another interface to easily add and remove data provider authentications from own profile. It's simply much easier for all new small startups to have this taken care of by someone else instead of having to build the wheel all over again. And it's also better for the users if the user experience is similar each time.
3) The money issue. A single entity with API deals in place with all the major players could move things forward faster. Then the big sites didn't have to provide their streams totally free, they could take part of the revenues collected by the middleman. It's more of a hassle for everyone if every startup needs to make those deals. Obviously this is not a problem if big sites choose to provide their data and the needed OAuth apis for free.
Rebecca, I agree! Maybe the guys from TrustCloud hear my cry and start working on it. :)
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Will wrote on 01.10.12, 4:40pm:
If you rent your house you are taking a risk and are probably the best person to check references rather than trust a relatively new website (airbnb)
Hans Schoenburg wrote on 01.10.12, 8:03pm:
Juho,
You are dead-on here. I would go further and question whether the aggregation of reputation across platforms even generates important value. Reputation is context dependent. Being a good driver doesn't make me a good house-guest.
In my opinion, the quest to consolidate reputation into a singular score is a perceived need, a misguided attempt to recreate our bank accounts.
Of course, the functional interoperability of the sharing economy is of utmost importance. But reputation should remain plural.
Hans is a co-founder of GiftFlow.org, an online platform for supporting the real-world gift economy.
Sam Cain wrote on 01.11.12, 2:56am:
Incredible post Juho.
Would love to talk to you about TrustRank.me
We're trying to solve the exact problem you're talking about.
Best,
Juho Makkonen wrote on 01.11.12, 3:59am:
Hans, you are right. Airbnb, eBay, Zimrider and TaskRabbit all have different contexts, so it might not bring enough trust in one service to have a good reputation in another. But I think it would still help a lot. For example, if someone has slept in 10 houses and all hosts have given that person tremendous feedback, it would be an easier decision for me to lend my power drill to that person. Good reputation from 100 sales in eBay would also help me make the decision. A very high Klout score? I think it would be better than having zero internet footprint, but not that much.
Sam, thanks! Based on your website TrustRank seems to be doing almost exactly what I proposed! I knew I couldn't be the only one thinking this should be done. :) I submitted my email through your site, would love to talk with you about how you guys are going to pull the data from different services etc.
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Mauro Nunes wrote on 01.11.12, 5:03am:
Hi Juho, this post could not have come at a better time.
We have just launched WhyTrusted.com to do just what you wrote.
We currently integrate eBay, Facebook, LinkedIn and Twitter. Have a look and give it a try if you wish. Your feedback would be very welcome! :)
---
Mauro is a co-founder of WhyTrusted.com, a trust aggregation service based on a user's online reputation trail.
Juho Makkonen wrote on 01.11.12, 5:38am:
Mauro, thanks, sounds very interesting, I'll check your service out!
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Chuck Chesler wrote on 01.11.12, 7:56am:
Juho:
We agree completely. A good driver doesnt necessarily make a good house guest. All we can do is try and tip the odds in our favor as no prediction could ever be 100% Ultimately the best way to predict trust will be a combination of hard data ( verified address,phone number, credit score etc) combined with behavioral data ( social media activity, purchasing habits, etc). That's what we're striving to do here at Trustcloud.
Chuck
Juho Makkonen wrote on 01.11.12, 8:32am:
Mauro, I joined your service and tried it out a bit. So at least eBay has a way to pull the reputation data from their system with OAuth, for free, and there are services that are doing it already, like you guys, that's good to know!
I like the fact that you show which data you pull from different services. However, I feel that the "trust score" you are counting somehow algorithmically is a bit ambiguous. It doesn't really tell me anything. It now says that I have a "good trust score" with value 2.09 because you recommended me, I have one positive feedback in eBay and I have 404 friends in Facebook. I'd suggest you to move in a bit more transparent direction. Also, I believe Facebook and eBay are too different as contexts, so you shouldn't try to create an algorithm that uses data from both to calculate a score. I presented an alternative solution in my article, what do you think about it?
Chuck, thanks for your comment. Using hard data (verifying address, phone number, credit score etc) is obviously an interesting field too, while it obviously also has some problems: phone numbers can be from prepaid subscriptions, address verification is a tedious process, and all this is really difficult to scale to different countries.
Chuck, I'd like to hear more about why you have chosen to present the data in the algoritmic way you are currently using and hide the data under the simplified badges. To me it feels too ambiguous. Why not use a simpler, more transparent system? What do you think of the methods I presented in the article, and the criticism I pointed towards you? I'd love to hear counter arguments. :) Also, are you guys planning to integrate with eBay etc.? Anyway, best of luck to you guys, you're working on a problem that's really important!
So there are now at least four players that I know of that are trying to do something that is pretty close to what I described: TrustCloud, TrustRank, WhyTrusted and Legit. I'm pretty sure there are dozens of more. I'd love to hear from them!
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Juho Makkonen wrote on 01.11.12, 9:16am:
Mauro, I'm just kind of wondering why you have to have the score in the first place, since it can be a bit misleading. :) Why not simply have aggregate of all reviews and/or amount of positive ratings as the score, if there needs to be such thing. I do understand that it's tempting to have a convenient single figure, but I'm not sure whether trust can be reduced into a single figure in such a way. It's non-transparent and simply confusing.
Anyway, it really seems that you guys have already implemented a lot of what I was going after! So your plan is to allow sites to implement the "trust card" into their services and pull data from your site to their profiles? Do you have any sites using your system yet? And if you guys are a startup, as I presume, how do you plan making money with your service?
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Sam Greenlee wrote on 01.11.12, 9:53am:
Great points, Juho. I think another point that your post moves toward is the necessity of abandoning anonymity for building trust in sharing. While anonymity and privacy are important values that should be protected on the internet, it seems reasonable that at least in the realm of sharing real things people must be willing to flesh out who they are.
An aggregating middle-man could, therefore, also encourage users to treat their profiles similarly to about.me, meaning that they are encouraged to link to many instances of their internet presence. By offering trust scores like you discuss alongside a person's blog, employer, school, Facebook profile, Google+ account, Twitter handle, etc., people are less likely to mistreat others' possessions (since they know that they are not an anonymous person).
Also, I think that the grand-scale projects of cooperating over huge distances (like airBnB) will increasingly rely on reputation established by sharing at the local and small-scale level where people actually know one another.
Mauro Nunes wrote on 01.11.12, 10:03am:
Juho, our trust score is just a way for someone to say, “Hey! In short, this is why I can be trusted, if you still have doubts, I can send you a trust token or you can check each rep source in my public profile”. Nothing to hide, just pure and simple reputation trail.
To be honest, we just wanted to do something meaningful and challenging related to collaborative consumption. We have a lot of ideas to expand the service and we are looking for as much feedback as possible.
---
Mauro is a co-founder of WhyTrusted.com, a trust aggregation service based on a user's online reputation trail.
Becca Smith wrote on 01.11.12, 10:39am:
Hi Juho, Becca from TrustCloud here -never fear, we hear your cry! We have recently been revamping our product to encompass some of the important elements you mentioned. Part of our blog about Klout was about identifying why there was this sudden backlash towards them. It is definitely an issue of transparency, and that is something we will be addressing with our product moving forward. We are adding layers to TrustCards to make sure the score takes into account your interactions within the sharing economy. Partnerships with other start-ups are going to be key for us. The more platforms that integrate with TrustCloud, the better the TrustCards will be. Look out for some exciting announcements from us along with our new and improved website coming soon. Great article, and thoughtful comments by all.
Becca Smith
becca@trustcloud.com
www.trustcloud.com
Mauro Nunes wrote on 01.11.12, 10:55am:
Juho,
If I understood correctly, your solution is:
“In their users' reputation cards, two things are shown: the number of ratings and percentage of positive feedback. That's all the information a trust widget needs to contain.”
If it is, than we have it done already, look at my or your public profile (whytrusted.com/mauro.nunes ). We present a summary of your reputation trail and a link to each source profile.
Our trust score is an algorithm inference based on more variables than just the user’s data. I do agree that it is just a perception of trust, but would you not do business with someone with an excellent track record on eBay, has a social life (Facebook), does something important in LinkedIn and is interesting enough to have a lot of followers on Twitter?
Thanks for your time and good feedback.
---
Mauro is a co-founder of WhyTrusted.com, a trust aggregation service based on a user's online reputation trail.
cameron tonkinwise wrote on 01.11.12, 11:20am:
I think that there are some issues outside of trust that are crucial to the trustworthiness of a sharing relation, but being outside are un-metric-able and untransferrable.
For instance, if I am really attached to a product, the level of trust I need to share that product will be much higher. Clearly, there is also a threshold here: there are people so fetishistic about their possessions that no amount of trust will ever be enough. And those in the sharing economy movement should probably stop wasting their time trying to get those materialists to share.
On the other hand, people who care very little for material possessions might be trustworthy (because of their other values), but not for the sharing of goods. You want the person you are going to be sharing with to be the sort of person who will take care of goods.
And this points to another issue: that assessments of trust are not just assessments of values, but also of skills. If I trust you with my power tool, I am not just trusting that you would like to take care of it, but also that you have the technical capacity to take care of it, to use it in the right way.
The big thing that is missing in all this discussion of trust metrics is the difference it makes when the sharing is commercial vs social. There are very different levels and types of trust involved in:
a) someone wanting to make some extra cash by lending possessions they are not currently using
b) someone happy to lend possessions they are not using but wanting some money to be involved just to make the transaction feel more secure or convenient (e.g., knowing the borrower will show on time, and return on time)
c) someone happy to lend possessions without a bond
This why there is an enormous difference between couchsurfing and AirBnB. It is also the difference between:
- an AirBnB active host (making meals, acting as a guide)
- an AirBnB host in the same home/apartment but who is less interactive, and
- an AirBnB absent host giving you the keys to an apartment. As AirBnB gets more and more regular Vacation Rentals on its books, I worry it is headed for a Qwikster-like fiasco.
There is also then the question of whether you are sharing with an individual or a corporation (which in the US of course is considered to be an individual). Ebay ratings for me get muddled when the same rating scheme is used on some individual seller and some small business trading through Ebay. Again, the levels and nature of trust are completely different.
Finally there is the recursive problem. When you read a review of a restaurant on Yelp or Amazon, you always need to check out the reviewer's history of reviews, to see if they are a serial complainer, or part of some astroturf campaign trying to badmouth those particular kinds of books. How do you establish the trustworthiness of the trust rating agencies? Are they for-profit, in which case is the business model transparent? What if they are part-owned by banks peddling credit card fraud insurance, etc?
Some of these points are in a tedious draft report available at: http://dasaufnahme.wordpress.com/publications/
Cameron
Steven Poe wrote on 01.11.12, 12:59pm:
@Mauro - I like where WhyTrusted is going. I think social and transaction rating/score should be visually separate. Providing the links for further inquiry about the person is excellent.
There are plenty of people who refuse to become a product on Facebook, or other social media plays, who have impecable digital footprints of tangible transactions, which are based on facts.
Social reputation or popularity is murky in that it's easily bought online (fiverr), and charismatic social personalities can't be attributed to trustworthiness.
@Cameron - thank you for thinking about and sharing this great piece. Breaking out the different levels is a great addition to understanding the share economy trust piece.
======
Creative technology founder pondering collaborative consumption...
Juho Makkonen wrote on 01.12.12, 2:30am:
Cameron, thanks a lot for your insightful comments! I have to take a look to the report, thanks for sharing. Others, thanks a lot for an interesting conversation, it really seems that a lot of people are thinking about these issues and things are really moving forward.
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
fla wrote on 01.27.12, 5:13pm:
wow, thanks for sharing to all.
i can trust you cos you have lot of activity in privative networks (regardless if i can pull data out), but you couldn't trust me because i do not use them because they are privative or will you trust me more cos i don't use them? (at least i will 'generally' like-trust you more if you don't use them)... so trusting you what for if i don't like your nonsharing producing habits? (see also: trust rating agencies information 'sharing') but yes, the more you like your fb profile, the more you will be afraid of it being shited.
/the general person's values issue/ {..too generalistic anyway..?} 'not too privative' people (i.e. a forever free lover) are more transparent generally, so more trustworthy in what they say you.. }
..further than trust, i prefer to share with fetishists of x.. in the abundant end.
..fair (distributed) insurance network(s) could help meanwhile..
i used to used couchsurfing but since long time i prefer the tramp way looking for people feeling the global freedom in themselves. i have a good scan for that.. cos i worked it out myself vanguardistlessly.
i am also looking for that "number-qualification", but the algorithm to be made for a transparent program of (tending to) distributed currencies for avoiding lots of other numbers in its interfaces.. that will be type of insurance networks wished i guess.
besides of that:
..we also need to get onto all using a 'same' sharing license, this way things are more clear from the begining.
And later on the more to share, the less to 'generally' fear about further sharing later and a blacklist threat if you behave bad could discourage your free riding option more, so less need of certified grants given in advance.
and in that license simple says 'some hard data before please', and 'if you run away i could pass it on' (to a global blacklist?) .. . Hard data (weird example: what do you love the most?) make sense to be pushed forward as the main reference for trust along grants in fungible or specific goods of my interest cos sharing is tending to be local-not corporate nor transnational where hard data could have more importance..
i have another idea, but it's too much already, :P
my manifesto-resume: for the sharing economy, trust ratings and grants' networks should be based on shareable applications producing content in a sharing direction, it will light up all the rest of trust bureaucracy and we won't get that far in sharing otherwise.
-------
i just stumbled upon this article and felt to dump this, poster's last comment looks like he was trying to say 'see you later around people'.. so
hi! :p
i edit lots at http://sharewiki.org
Charles H. Green wrote on 03.07.12, 5:19pm:
The problem with algorithmic approaches to reputation is not the algorithm - it's the reliance on reputation.
Trust is not reputation. Klout does not measure trust. This article continually glosses over the difference, implying (or so it seems to me) they are the same. They are not. Let me break it down.
Reputation is what people say about you. Trust (actually, trustworthiness – let's be clear, we're not talking about propensity of a trustor to trust, but of a trustee to be trustworthy). is only partly what people say about you. It is also, very much, about what you do.
When Edelman PR tried its TweetLevel tool, it had four components: one was trust. Trouble was, it was a reputational algorithm, and Justin Bieber kept winning it. And not because we would describe him as trustworthy.
If I make 25 appointments on a social network, and show up on time, as contracted, I would call that reliability – which intuitively we think of as part of trust. Which data-set would you prefer to have to gauge my trustworthiness – my 25 behavioral actions? Or 300 of my bff's who I've managed to get to vouch for my reliability? I'll take behavioral data any day of the week.
One reason behavior beats reputation is that reputation is easy to game; and the simpler you make the algorithm, the easier it is to game. Making "funds of funds" out of many algorithms is no different than packaging lots of low-grade mortgages into a mortgage-backed security; the data is only as good as the underlying data.
By contrast, behavior reveals character. Aristotle said it: excellence is but a habit. For me to game my 25 show-up-on-time appointments, I have to show up on time 25 times. After a while, unless I'm building a major sting operation, my data is my life, and my life is my data. No faking it.
Another reason reputation is only weakly associated with trust is that it is only weakly transitive. I use LinkedIn as a resume source – I'm not really going to refer anyone who's three degrees away from me, not for anything important.
The problem is not the algorithm, it's the reliance on reputation. Trust is even more contextual than reputation. In my own work over the last 15 years, I developed the Trust Equation: a statement of four factors of personal trustworthiness that cover most common-language uses of the word. The four variables are:
Credibility: basically can I trust what he says (track record, credentials, logic, etcl)
Reliability: can I trust what he does - consistency, integrity, meets promises
Intimacy or safety: can I share information or secrets with this person
Other-Orientation: are the selfish and neurotic, or do they play nicely in the sandbox.
The right road to a valid trust metric is to take that proven model (25,000 people have taken a self-assessment test based on it*) and match it up to behavioral metrics. I have to disagree with the authors here about social media as data sources, I think they are rich fields for mining.
So at the risk of repetition: look for behavior not reputation; and base algorithms on those behaviors.
*If anyone's interested in the self-assessment based on the Trust Equation, it's called the Trust Quotient (TQ) and you can take it for free at http://trustsuite.trustedadvisor.com/
Juho Makkonen wrote on 03.08.12, 2:42am:
Hey Charles, thanks for your thoughts. I completely agree with you on that trust is based on behavior. The system I propose does exactly that: it simply measures the behavior of people, based on testimonials by others. I think that's a great way to build trust. Obviously trust, reputation and behavior are not the same things, but they are closely related. Basicly reputation, as I see it, is an expectance of a person's future behavior based on his behavior history.
I don't think that a having a huge amount of friends, followers or retweets tells that much about your expected behavior, that's why I was criticizing Klout. However, if I have a good friend who I trust, and that person is friends with someone I don't know, that maked me trust the said person more than if he was just a random person I know nothing about.
I'd like to know more about your ideas on how to measure behavior and build trust by looking at data from social networks. I do think there are interesting possibilities, but also problems, like the one about Justin Bieber you mentioned.
--
Juho is a Co-Founder and CEO of Sharetribe, a platform for creating custom peer-to-peer marketplaces. http://www.sharetribe.com
Charles H. Green wrote on 03.09.12, 7:42am:
Juho,
Thanks for your prompt response back to me.
I would say that measuring the behavior of people based on the testimonials of others is not the same as directly measuring their behavior.
That's why we have rules of evidence in court that rule out hearsay. That's why writers and reporters work so hard to strip out opinion from fact in the people they talk to. That's why editorial and reporting are separate in newspapers. That's why Hemingway eschewed adjectives.
If I make 25 appointments and keep 24 of them, then in theory 24 people will report I was on time and one will say I wasn't. But that's never what happens.
What happens is half those people will assert "Charlie is always on time," despite having only one data point. What happens is the guy whose appointment is missed may have a Klout rating ten times that of everyone else, and he says, "Charlie is always late." What happens is that two of those 25 will confuse me with someone else and attribute my on-time behavior to another Charlie. What happens is that another couple of people confuse me with always-late-Christopher, and I get blamed for not showing up at appointments that were never made by me.
The point is simply, reputation is not trust. When using reputation as a proxy for behavioral trust, you're not only taking the data one step removed, but allowing all kinds of human subjective perceptual bias into the equation.
Testimonials of others definitely have their place. For example, there are aspects of trust that are purely about interactions ("I found him very easy to do business with"). In those cases, the testimony itself is the evidence – it is data of the first order, not the second, it is data about the observer rather than the observed.
The closer the relationship and the fewer the degrees of freedom, the more we place weight on testimonials. As you suggested, a friend's recommendation of a friend carries some weight. But that friend's recommendation, in turn, of another friend? Suddenly the transitive effect of friendship is shown to be very weak.
And if you measure thousands of people talking about the "reputation" of someone, what you have is a measure of popularity, or notoriety, or visibility; some kind of iconic opinion-leading power. That's very important to marketers and pollsters to know that kind of metric--after all, that can sell records and win elections. But that is different, I think, from trust. To deal with trust at the higher levels of scale, behavioral data becomes much more desirable.
Thanks Juho, good points for issues I've been wondering about.
I have to ask this although I clearly have an axe to grind: How do you see the role of "frictionless sharing" in constructing more reliable and transparent reputation systems?