Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: s.jackson on October 11, 2013, 03:16:05 pm
-
Hi everyone. I'm new here :)
I'm currently working on a project I find very hard to do just by my own.
Predicting Future Events using text mining and datasets such as GDELT, with AI
Does anyone have any idea how can I start? Or maybe there have been similar works before?
Thanks :)
-
Welcome!
It's certainly the rage of the age with Google, Amazon and other companies practically knowing our every move before we do. It's almost scary in a perverse and pervasive sort of way. The late Janice Joplin said it best, "Freedom's just another word for nothing left to loose." - (Me and Bobby McGee).
We practically had no freedom left along with morality, manners, shame, etc.
But hey, I didn't mean to rain on your parade as I'm sure your intentions are honorable (right?) ;)
While we might be able to offer some insight here, do check out www.chatbots.org (http://www.chatbots.org) as they have some members who eat this sort of thing for breakfast.
Hopefully someone here can /will chime in on your quest.
Best of luck and again welcome!
-
Something came to mind like this a few years back where a program was claimed to be predicting events. It might be this I am thinking about :
http://en.wikipedia.org/wiki/Web_Bot (http://en.wikipedia.org/wiki/Web_Bot)
There's also a little article here that might interest :
http://www.theguardian.com/commentisfree/2011/sep/11/charliue-brooker-computer-predicts-future (http://www.theguardian.com/commentisfree/2011/sep/11/charliue-brooker-computer-predicts-future)
I don't think either is what I am thinking about but hey, maybe some research may turn up others.
-
That's a tough nut to crack indeed. Here are some tips that could help you on your way:
-from my kaggle experiences, I have learned that n-grams are very well suited as a first step for processing text in order to do predictions.
-you're probably going to need a lot of training data, so the GDELT data set will probably be very helpful (although I don't know this data set personally). Usually, you split up these data set in 2: about 66% for training your algorithms, the rest to use for verification.
-Some good algorithms to consider: decision trees (and it's simpler cousin the decision tree), Bayesian neural nets (although these can be very hard), naive bayes classifiers might also be helpful. I haven't yet worked with hidden markov models so no idea of they can help,...
-In general, I find it very hard to predict which algorithm will give the best results, so things probably need testing.
-scikit is a very popular python based library that already implements a lot of machine learning algorithms, so that's a nice place to start experimenting.
-
What kinda Future Events prediction have you in mind? I mean, predicting where user going during conversation with bot or just like woodo magic of oracle? ;-) First can be done to quite extend but other one.....hehe I just dunno ..smile.... maybe GOD can only tell. So resurecting your religious beliefs might be good idea. Who knows, GOD makes you the choosen one..... ;-)