Although this is probably less research-y than usual posts to this list but I thought the use of ML to analyze a community curated corpus of positive and negative online behavior was interesting: