6Tribes was a social network based purely around interest groups. It would read from your Facebook data as well as phone music and location, and try and identify what you like. Here I worked as a Data Scientist on two key products, being the matching of users to interest groups, as well as Social DNA.
Rather than focusing on DNA being your biological and genetical structure, we wanted to convey what you consist of as a person through your personality, likes and disliks: What do you like/hate? Where do you hangout? This product allowed us to autofill a users profile, with a series of hashtags/dna strings which they could then later remove and adjust if needs be.
Matching to Interests Groups
Here we matched users to relevant interest groups. On other sites, such as Twitter it is more about connecting and following someone you admire. But how would you follow a set of people with the same interests? We made a very controversial and bold decision showing the user a match percentage. We wanted the user to get excited when they saw higher matches and for this to resonate with them.
Usually sites and apps will get you to invite people via email or direct app specific invite. How ever we wanted to try increase precision by suggesting friends whom we believed and inferred held a connected interest to you. Hence a invite would go to a user whom we believed would also be interested in Group X.
At first, the above may seem like a straight forward problem. How ever we had two key difficulties: 1. We had no data on the user side, nor on the groupings side. 2) Facebook data such as likes is extremely old, matured and outdated so does not represent the user, so how could we increase accuracy? To handle this, we dealt with natural language extraction from users status updates, tagged photos places along with iPhone data. This was enriched through using Alchemy's API when we obtained a shared URL, along with Foursquare and Maps to identify types of places. Facebook provides very little URL detail and very bad categorisation on other items such as places, as we wanted to understand more than the level 'Restaurant'.
Engineering wise, we made the Data Science platform a microservice, using Flask, along with Circle CI and AWS for fast deployement. We built our algorithms so changes could make an impact within a minute, so we were responding easily to user feedback.
The issue with 6Tribes, was that at one point it was focused on being a musically oriented social network, then another time for community groups. Unfortunately it lacked direction. How ever it did lead on to become DriveTribe, which is the associated app to The Grand Tour (aka. Jeremy Clarkson's Top Gear).