1. I would look at a corpus of twitter data from two or three different points of the site's history. [Here](https://github.com/shaypal5/awesome-twitter-data) is an example of where I could find twitter corpora, and I believe nltk also has a way of parsing twitter data. I would probably want a very varied and large sample so as not to be biased to a particular twitter community. Or maybe I want to focus on a specific community and do the sentiment analysis for them? That would reduce the size.
My goal with analyzing this data is to see if online culture, specifically twitter culture, has gotten more negative as time has gone on. People often say that social media is a plague, but I have seen a lot of productive discussions to contrast against any percieved negativity as people realize the potential of being this globally connected. I want to test the hypothesis that social media both "is more negative", and "is more productive" and see the relationship they have to eachother, if any. For negativity I will look for words with a negative connotation, and for productivity i will look for social and political and scientific jargon and see how frequently it is being used.