More NLP-related ideas for these data

[abenton]

• Could try extracting more NLP-ish features beyond just n-grams. e.g., use OpenIE to identify predicates/arguments in tweets:
  • https://github.com/hltcoe/PredPatt
  • http://reverb.cs.washington.edu/
• Does creating a "story" across multiple tweets help increase follower count? This might be established by topical coherence, long coreference chains, or back-and-forth DMs. In general I like the idea of trying to cluster multiple tweets together into coherent blocks and analyzing the blocks. This is hard in general (which is why most Twitterology analyzes one tweet at a time), but may be something we can do since we've been following these bloggers for quite a long time.

[bellecarrell]

Does creating a "story" across multiple tweets help increase follower count? This might be established by topical coherence

I think this goes along with the hypothesis that "staying consistent" leads to more followers. Do you think this could be established using the features you've already added related to topic?