Local Discovery: Designing The Next-Gen Recommendation Engine | Hoppit

This post is dedicated to Hoppit, a new social place recommendation engine that recently launched in NYC and Chicago. Hoppit fosters local discovery - defined as the discovery of new restaurants, bars, shops, and activities around you. Throughout the past few years, I've seen interesting shifts in the local game. However, none have hit the nail on the head. There was still something missing.

Let's talk about Yelp, then Hunch, then Bizzy, then Google + Alfred + Ness. In that order.

Yelp got out there first, of course, providing the aggregated star ratings of local places. This directed people to good places, but Yelp didn't really provide any recommendation engine. It didn't return unique results for unique people. So, those people looked elsewhere.

Then, Hunch attempted place discovery - even though it was part of a much more grandiose mission of personalizing the Internet. The Hunch taste-based method said a lot about local innovation. They were the first to consider, "Should I trust people just like me? Can I base an recommendation algorithm off of human taste and psychographic data?" Definitely intriguing, but it misses a vital component - the actual attributes of the places themselves. We can't do taste-matching alone, because then we miss all this valuable place data.

Then came Bizzy, with an interesting thought. "Hey, if two people like the same place, then they're a good taste match." Again, interesting thought, but misses the boat. Two people can like a place for very different reasons (i.e. ambiance vs. food).

Finally, the market has been met with a slew of new "gene-based," Pandora-like place discovery players. Specifically, Google Places (formerly Hotpot), Alfred and Ness have built algorithms designed to isolate attributes of places. Then, if you "like" a place (or 5-star a place), they can spit back another place with very similar traits. Now then, this is the most compelling algorithmic approach yet; they're getting really close to the Nirvana of local discovery. BUT, it's still missing something.

It's missing what Hunch had. The taste-mapping element.

Gene-based approaches will never optimize place discovery, because everyone who 5-stars a restaurant will get the same results. It's still not tailored for each unique user.

By combining consumer taste-mapping with a gene-based approach, we can now sense which unique human tastes will like which unique place genes. And that is how Hoppit is designed. When a person specifies their tastes and psychographics, Hoppit immediately knows which places to point them to. Even more, the moment that same person likes (or 5-stars) a place - or several places - Hoppit can further customize its database of places for that user, based on the attributes of the places rated.

It's our belief that the best algorithmic approach to local discovery needs to encompass both taste mapping and place genes. Only then can we drive the right person to the right place at the right time.

Meet Hoppit.