[Feature Request] Extract address from text?

[valerieashborne]

I often found myself seeing many messages from different chat programs about some kind of events, meetings, etc. And they often list one (in most cases) or several (rarely), like:

"Hey, today at 18:00 will be this cool event! Everyone is welcome!
It will be at:

Oranienburger Str. 5050505676
0777643 Berlin

See you soon!"

And it will be nice to extract address from text and maybe even try to open/search for this address in map program (google maps, comaps, etc).

Some things to consider:

• language/region/country specific formats. Not all addresses are written in the same way, so some research is needed.
• Even with that, some people might prefer to write addresses differently then what country might consider as standard.

[corphish]

To add support for this, we will need to do entity extraction on texts, NLP comes into picture. Luckily, we do have a library from Google for this - MLKit. However, the results I got did not really give me confidence in adding this feature.

I tried on the text you mentioned above and it considered the "5050505676" as phone number and did not recognize any address at all. So, I would wait a bit to see if they improved before considering to add this feature (although this lib has been in beta since ages). Meanwhile, I will check if there are any alternatives.

[boognish-rising]

Edit - #30

this idea got me thinking, so I'm piggybacking off of it 100%:

It might be a bit easier/doable to extract http/https URL addresses. I know that's not what OP is asking for but they can take credit for the inspiration lol

So in effect, if one were to highlight plain text that contained non-hyperlinked URLs, ideally, the function would scrape those and provide them without the rest of the filler.

It's not something I find myself needing often, but definitely could have benefited from this type of functionality on several occasions in the past

[valerieashborne]

This warrants its own feature request, I think. There are many interesting things that worth being extracted

[valerieashborne]

To add support for this, we will need to do entity extraction on texts, NLP comes into picture. Luckily, we do have a library from Google for this - MLKit. However, the results I got did not really give me confidence in adding this feature.

Does good old regex could work?
Maybe somebody already crafted address extracting regexs somewhere deep in stackoverflow forums?

We do not need to parse HTML with regex after all, so should be doable

[boognish-rising]

This warrants its own feature request, I think. There are many interesting things that worth being extracted

You're right. Done #30