Thread by @asanwal

One of the hardest challenges we've had @cbinsights is building an algorithm to identify competitors

We can augment and deliver algorithmic recommendations to an internal team who will curate before clients see it

2/ But building an algorithm that perfectly does this without human assistance is one of those problems that bothers me in my more core as I think it should be solvable

But the problem is competitor relationships violate transitivity

3/ So if A is a competitor of B and B is a competitor of C, A is not necessarily a competitor of C

Lyft is a competitor of Uber and Uber is a competitor of Grubhub but Lyft and Grubhub are not competitors

The algo we built looks at this, co-mentions in media and a # of factors

4/ Our algo is more sophisticated anything else out there, BUT the signal isn't perfect so we can't go right to production with the recs

If you find this problem interesting or have solved similar challenges, I'd love to chat

We're also hiring cbinsights.com/jobs

</rant>

5/ btw, violating transitivity is not the only problem

Sometimes, company A considers company B as a competitor, but company B doesn’t consider A a competitor

These types of asymmetric competitive relationships make this an even more interesting (and difficult) problem

6/ Penned a post detailing the challenges of building this competitor algorithm here. Welcome thoughts esp from engineers who've solved similar problems or who have hypotheses on how to solve cbinsights.com/research/team-…