Crossing the 50,000 Place Line in Jakarta

StreetCred Team - May 13, 2020

Two months ago, we launched MapJakarta, challenging local players to win Bitcoin by filling in a blank map of their city. As of this week, those players have added 50,000 places to the map, a huge milestone for our mission of mapping the Indonesian capital.

"Many companies have spent money to collect this type of data once, but when they're done, it quickly goes stale. For example, businesses close, hours change, et cetera," StreetCred CEO Randy Meech told a reporter for Nikkei Asian Review in a recent interview. "Crowdsourcing POI data with an engaged community [ensures] that data can be collected and, critically, also be kept up to date."

All over the place

As the competition heats up, we’re seeing some players spread out across the whole city, while others dig deeper into their neighborhoods. Nearly all Jakarta game tiles in play have been at least 50% mapped, according to our POI estimates, and two-thirds have been at least 80% mapped.

Not only are players adding a ton of new places, they’re also bringing fresh eyes to existing ones. Our model automatically weighs data freshness, meaning the chance a place has closed or an attribute has changed since the last update. The typical place on our map has been updated in the past 15 days, meaning that data quality remains consistent over time.

Places in Jakarta
Nearly all Jakarta game tiles in play have been at least 50% mapped.

As the map becomes both more populated and more accurate, we’re seeing user engagement continue to surge. Last week alone, MapJakarta participants added and updated 21,973 places, maintaining the steady climb recorded since week one.

Notably, that increase was driven by a handful of top players. Every time you add or update a place, you have the option to drop a cube; at the end of the week, the players who dropped the most cubes score Bitcoin. During the most recent contest, each player in the top eight dropped more than a thousand cubes.

As new players discover the game and figure out new strategies for efficient mapping, we expect that number to keep climbing. There’s also a repeat game dynamic at play: every time players win a contest, they score additional cubes that they can use during future weeks.


Dealing with spam

One of the most frequent questions we get is: How do you deal with bad data?

Any contest admin must be prepared for spammers. Sure enough, we’ve seen our fair share of fake photos trying to masquerade as place data. However, we’re confident in the probabilistic modeling approach we’ve developed, which automatically identifies and excludes data from users with historically poor or sloppy performance. As our probabilistic model encounters more real-world data, it gets better and better at pinpointing what’s accurate.

And while we have seen a handful of repeat offenders, easy to identify by their consistent use of low-quality data, those users represent the minority. Looking at the typical place added by players in Jakarta, the data is 86% accurate. This number might sound high, but it makes sense given that StreetCred power users are incredibly prolific. Most of our data is contributed by users whose accuracy exceeds 90% after being vetted across dozens of trials. By crowdsourcing each data point, we can ensure that even incomplete data enriches our understanding of the POI landscape. As we continue to map the city’s bustling grid, we look forward to exploring Jakarta in even more detail.