Technology is at the heart of Traity. A group of brilliant Data scientists lock themselves everyday in our labs, in order to find out the most powerful and ethical algorithms to build trust between strangers online.
This is how the ReputationAPI, the real base of the Traity's infrastructure came to life. Marketplaces are able to push information and pull information (accepted by the user) in order to enter more reputation into the system and to integrate that reputation with their own profiles.
Then, what is the science behind our mission?
Reputation widget and API
First stage: Identity
We make sure that people are who they say they are, it is purely a matter of transparency. We ask users to login with different social networks and we'll check that those networks seem legit (in terms of friends, content, when they were created, common friends, etc.). We'll also go offline, to ask users to upload their passport or ID. We delete the picture, we just maintain the fact that "this user verified herself with the right name".
Second stage: Social networks
There are here more algorithms we are running. Who are your best friends? Where do you spend most of your time? Is it consistent with where your best friends spend time? If you say on Linkedin that you worked for McKinsey, do you have friends in your network who also worked for McKinsey? If they did not, it might seem strange. This does not cover all aspects of reputation but now, if you want to fake a profile, you have to fake 500 profiles of 500 friends, so we are making it more difficult for people to fake their social identities.
Third stage: Recommendations
When the user gives us access, we take reviews information from marketplaces. Algorithms here don't play a big part. Aggregation is more important. If you think about it, eBay gives people the % rating and the number of transactions. The % is important (you want to see 99.9%) but the number of transactions as well. Less transactions is worse, more transactions is safer. Even if you see 98% in 50 transactions, you will want to read "the bad one," so that's why algorithms are not so important here. It's all about transparency and letting people have access to the full history so that they can read the bad one, or whatever they want to make up in their minds.
Fraud plays an important role in the world of insurance. An example is that there are more claims for the theft of a Rolex watch than the amount manufactured worldwide. Therefore, it is important to know that the type of people who get insurance are more trustworthy. To do so, we go to reliable indicators to show how a person has behaved in the past, for example, their reviews, to conclude the risk index of defrauding the insurer. We have a clear approach versus the traditional practice of insurers: everyone should not pay for the fraud committed by only a few people. That’s why we offer personalized insurance based on their reputation. This was developed by using machine learning, state of the art technique, such as deep learning, and social network analysis.