Address-Quality

In the previous post, I asked why address forms still make you select information you have already typed. My proposed solution was to move that effort into an API. You provide the address, the API interprets it, then you confirm the result. So I built Address Quality, an experiment for turning free-text Indonesian addresses into structured administrative data. The first alpha version tries to answer two questions: Which parts of the address can it recognize? Which possible location matches those parts best? That sounds simple. But something that sounds simple isn’t always easy to solve. ...

One question keeps coming up while I build Address Quality: Why not just use Google Maps API? If I have a messy address, Google Maps API can turn it into a structured result. It also knows roads, buildings, businesses, and landmarks that my administrative data doesn’t. It is the top of mind when we are trying to solve an address problem. Take this place name: Paskal Hyper Square On its own, that doesn’t tell me much about province, city, district, or subdistrict. Yet, Google can still make sense of it. ...

I’ve noticed something common about address forms in Indonesia. You type your address. Then you have to fill in other fields such as kelurahan(subdistrict), kecamatan(district), city, and even province. No matter how complete your address is. For example, you have entered this address: Sudirman Central Business District(SCBD), Jl. Jend. Sudirman kav 52-53 No.LOT 6-8, RT.5/RW.1, Senayan, Kec. Kebayoran. Baru, Kota Jakarta Selatan, Daerah Khusus Ibukota Jakarta 12190 Complete and well structured. But then you have to select again: ...

This is the first post in a series about building Address Quality, an API that validates Indonesian addresses. I want to document it honestly: the failures, the measurements, and the changes in thinking. It is not a “nailed it” story yet, because the work is just getting started. This post covers the why and the how. It ends with the first benchmark: 49% accuracy on my own test set. I will explain later why that number turned out to be useful. ...