> That type of data would be so interesting for things like historical sentiment analysis

Except that internet commenters are very weird, and like the least representative sample ever. Not quite as weird as Wikipedia editors, but still really weird.

They aren't representative of the general public. This can still make it very interesting though. Do trends show up earlier among commentators? If so, has the time it takes for the trends to flow to the mainstream changed over time. Has the likelihood at which online commentator trends flow to the mainstream changed? It's the influence more pronounced for specific subjects?