So courtesy of Covid-19 I had some time on my hands. I went back to some earlier abandoned projects of mine that I had started some years ago.
One of them was @poweralertsbot. A social media powered bot that searches for and catalogs power outages in Kenya. It uses machine learning to find tweets about power outages and probable locations that were mentioned. It has been cataloging power outages since 2017.
It would likely reply to you faster that you’d get feedback from KenyaPower.
Some interesting things I discovered.
Kenyans have terrible grammar compilation.
In trying to train the data model to find tweets / posts that were actualy power outages. Kenyans for the love of their mother enjoy shortening simple sentences to what they seem fit. And we haven’t even taken Sheng into consideration.
Kenyans tweet aloooooot about power outages (roughly 409,517 tweets on average each year )
In between when I started this project, 2017-04-17 and now, they’ve been nearly 1,228,551 total tweets and retweets referencing either power outages, kenya power, or blackouts.
Nairobi isn’t the area with the top number of power outages reported.
Interestingly enough, Kisumu takes that lead. By a small margin though. Take into consideration however, that the number of outages reported and the number of outages on the ground are not necessarily equal.
If you would like to find out outages in your area, simple tweet to the bot with the word status and a location and it should struggle hard enough to pick a location and reply to you.
Currently on my development plan is
- [ x ] Migrate the Machine Learning to Improve on semantics
- [ x ] Add a bot reply feature on Twitter
- [ ] Fix the android mobile app to pick from the new DB
I’ve lost my android developer key so this might be a while before this get’s done
- [ ] Add an API to allow other developers to piggy back off from the App.
Also, I’m keen on any feedback you guys might have with regards ti feature requests as well as data analysis.