As somebody coming from geocoding background parsing places from full text will be a huge challenge. Hardly a word that isn't also a location. (There's a state in Turkey called Batman).
Indeed it's one of the most difficult challenges I am facing, the solution would be to give the power to the users so that they can discard such parsing. But first I have to ensure that there are people who might be interested in using the solution, it's the phase I am in right now (trying to validate the concept), thanks for the feedback :)
Re:Show HN: News geolocation website
by DyslexicAtheist on Monday July 23, 2018 @12:52PM
(#17593972)
there is a town in Austria called Fucking: https://www.youtube.com/watch?v=bbNzPnvkn-A
haha, true, indeed
> built almost exactly the same thing in 2009 (at the peak of 'hyperlocal' hype)
In 2009, there was around 25 million smart phone users. We will close out this decade close to 3 billion. That is half the world with a geo-location based device in their pocket which grabs their attention for hours each day.
The peak may have been Pokemon Go, or maybe we aren't even there yet.
We probably aren't there yet in practice, but the hype far preceded things like Pokemon Go. Foursquare really kicked it off.
true
True, there are many things to do to improve the quality of services/products using the devices geo-location, not only in the gaming business.
Thanks a lot for all the feedbacks, I agree with you, the more I work on the project the more I realize how difficult it is to make everything fully automated without impacting the quality of the geolocation (NLP is not 100% reliable) and the quality of the sources, I am still working on that.
1. Your biggest problem is going to be acquiring audience; 'if you build it, they will come' is probably not going to play out unless you offer something above and beyond a typical aggregation web site.
2. Consider the utility of granular location-specific news. For many users what you will be showing is either very little content (because there isn't any) or content from a larger geographic region. At that point you have to ask who your audience is and what you provide to them over a bigger regional news source.
3. Keeping up with available sources will be a big challenge. Remember, if you're just aggregating from the big sources, you're not providing much; the appeal to an aggregator is that it gets and filters everything. If you have to go to another source as a user, there's no point in using this site.
4. The site looks pretty bare-bones. It could use some design help across the board.
5. The 'secret sauce' of a site like this is actually not aggregation and automation ... it's curation. If you read 100,000 RSS feeds (if and where you can even find them these days!) you still need to separate the wheat from the chaff. That's why Reddit works - it's effectively a user-curated aggregate site that leverages a scoring algorithm to (ostensibly) bring the best stuff to the top.
Edit: Forgot one of the biggest challenges. I don't know how you're doing your address extraction, but one of the toughest things we ran into were false positives. Articles that reference a location which we geocode but really weren't about that location. So you'll analyze a whole story (which can get very computationally expensive) and find 4 locations:
1600 Pennsylvania Avenue Washington, D.C. Washington Zimbabwe
Now this story is about the Zimbabwe ambassador having a meeting with the president. Two of those addresses are where the story is, two probably aren't (Washington would probably not be geocoded to D.C.). So you file the story in:
Washington, D.C. Washington, USA Zimbabwe
Now you've just got noise. NLP can help but it's not enough - you will mis-geolocate this story. When that happens a lot the results are not reliable and, again, your potential audience does not see the utility.
Anyway, best of luck.