um... make a 4chan app that lets users up/downvote threads, then builds a naive bayesian model of what keywords (eg. "toasting epic bread") are correllated with the kind of threads you like. Netflix-like. A sort of automatic cream-extractor.
de-anonymizer based on posting times, writing style, what baits them to respond, etc. The Thread-Local unique ID's would help, giving you more stuff that you knew came from same user. Don't know if this one is practical. Kindof scraping the bottom here...
track usages of phrases over time. (thinking of the recent evolution of "rustled my jimmies" derivatives)
See what topics are trending
Fuck maybe I should build some of this...