It’s not what software you use, it’s how you do frame the problem, define features with help from other resources, etc etc. As I said in Lesson 5.1, the actual data mining is just a small part of the problem! For software, you can certainly use Weka. If you are processing tens of millions of tweets you may find that the Explorer cannot load your full dataset, but there are ways around this (discussed in “More Data Mining with Weka”).
Just to whet your appetite, I attach a paper by a group from Chile (Felipe Bravo-Marquez is now at Waikato) that might help you think of new approaches — I really like it. The most interesting part of the approach is not in the data mining algorithms used. However, for your interest they used many methods you know about —J48, Naive Bayes, Logistic regression, and SVMs — and got best results from SVMs. Also, they used feature selection (lots about that in “More Data Mining with Weka”).
cheers
ian